A spiral bevel gear tooth flank is a conjugate surface, not a surface of revolution. Its geometry is generated by a rotating cutter head whose axis is tilted relative to the gear axis, and the final grinding pass must reproduce that geometry while leaving a roughness profile that can support an oil film and resist contact fatigue. The two requirements are coupled: the local curvature of the flank determines how each abrasive grain enters and leaves the surface, and the resulting grain traces determine the roughness, the bearing area, and the distribution of residual peaks.
This coupling is why surface-topography prediction for spiral bevel gears is harder than for flat or cylindrical grinding. The tool path is a three-dimensional trochoid on a curved surface; the workpiece rotates and feeds while the grinding wheel rotates; and in ultrasonic-assisted grinding a high-frequency, low-amplitude vibration is superimposed on the relative motion. The modeling chain described below connects the nominal tooth geometry, the acoustic design of the vibrating gear, the stochastic topography of the grinding wheel, the grain kinematics, and the final surface envelope.
1. The finishing problem in spiral bevel gears
Spiral bevel gears transmit motion between intersecting shafts with a gradual, overlapping contact pattern. They are used in helicopter tail-rotor drives, aeroengine accessory drives, and marine transmissions, where high torque density and low noise are required. The tooth flank is hardened before final grinding, so the grinding operation must remove a small amount of material without introducing thermal damage or tensile residual stress. Conventional grinding relies on continuous contact between the grinding wheel and the flank. The specific energy is high, the contact zone is small, and the risk of burning or grinding-wheel loading is significant.
Ultrasonic-assisted grinding superimposes a vibration, typically at 20 kHz and an amplitude of several micrometres, on the workpiece or the grinding wheel. The contact becomes intermittent. Each grain impacts the surface repeatedly, the effective cutting force falls, and the chip is removed in shorter segments. The surface texture changes from continuous grooves to a pattern of interrupted, overlapping traces. Reported effects are summarized in Table 1. The numbers are not universal—they depend on the material, the grinding wheel, and the vibration parameters—but they show the direction and the scale of the improvement.
| Case | Material and process | Ultrasonic condition | Reported outcome |
|---|---|---|---|
| 1 | SiC ceramic, ultrasonic face grinding | Amplitude 0–1 µm | Sa minimum 0.04 µm at 0.216 µm |
| 2 | Ultrasonic-assisted grinding | Amplitude 4 µm | Sa reduced ≥24%; contact stiffness +68%; peak contact pressure −17% |
| 3 | Single-crystal Ni-based superalloy, ultrasonic vibration-assisted grinding | — | Sa reduced ~29.6% |
| 4 | 45 steel, axial ultrasonic dry grinding | — | Ra and Rz reduced ~20% |
| 5 | Hypoid gear, ultrasonic lapping | — | Roughness reduced to 0.2 µm |
The practical question is not whether ultrasonic vibration can reduce roughness, but how the topography is generated on a specific flank under specific kinematic conditions. A predictive model must answer three questions: what is the undeformed grain path on the curved tooth surface, how does the vibration modify that path, and how do the paths of many grains combine into a surface?
2. Tooth flank geometry and the macro model
The geometric foundation is the local-conjugate method developed within the Gleason system and later generalized by the local-synthesis method. Instead of enforcing full conjugacy over the entire flank, a contact point and a contact ellipse are prescribed, and the machine settings are adjusted so that the curvatures of the large gear and pinion match at that point. At any contact point, the two surfaces share a common normal \(\mathbf{n}\) and satisfy the meshing condition
$$\mathbf{n}\cdot\mathbf{v}^{(12)}=0$$
where \(\mathbf{v}^{(12)}\) is the relative velocity between the cutter and the gear blank. The cutter surface is a cone. For the large gear, the cutter tip radius is
$$r_{02}=r_0\pm\frac{W_2}{2}$$
with the plus sign for the outer cutter that generates the concave flank and the minus sign for the inner cutter that generates the convex flank. The cutter surface is transformed into the gear frame by the machine-tool settings; the envelope of the cutter positions is the tooth flank. For the pinion, a modified roll with second- and third-order coefficients is added to control the transmission-error function and the contact pattern.
For the gear pair considered here, the blank parameters are listed in Table 2. The outer cone distance follows from the module and the tooth numbers:
$$R=\frac{m\sqrt{Z_1^2+Z_2^2}}{2}=106.94\ \mathrm{mm}$$
The face width is 30 mm, well below the outer cone distance. The large gear is right-handed. These values define the nominal geometry; they do not include manufacturing errors or heat-treatment distortion.
| Parameter | Value |
|---|---|
| Module | 3.6 mm |
| Teeth (pinion / large gear) | 41 / 43 |
| Shaft angle | 90° |
| Spiral angle | 25° |
| Pressure angle | 20° |
| Face width | 30 mm |
| Addendum coefficient | 0.887 |
| Clearance coefficient | 0.677 |
| Addendum | 3.196 mm |
| Dedendum | 3.6 mm |
| Hand (large gear) | Right |
| Outer cone distance | 106.94 mm (calculated) |
The continuous tooth surface is converted into a CAD model by rotary projection. The flank is projected onto the axial section, where four boundary curves are defined by the face cone, root cone, heel, and toe. The boundary curves are solved together with the tooth-surface equations, and the enclosed region is divided into a 5×9 grid for one tooth. Each grid point corresponds to a three-dimensional point on the flank. The point cloud is lofted into five curves per tooth, mirrored for the opposite flank, and arrayed around the axis to form the full gear. The result is a nominal solid model that serves as the substrate for the grinding simulation.
3. Designing a longitudinal ultrasonic horn around the gear
The large gear is stiff and lossy compared with a conventional ultrasonic horn. Forcing the whole gear–horn assembly into a conventional half-wave resonance would require a very long horn and would produce unwanted bending or torsional modes. The design therefore uses a non-resonant approach: the horn resonates, while the gear is treated as a coupled elastic load. The gear is simplified as a truncated cone with equivalent mass and stiffness, and the horn is represented as a stepped cylinder–cone–cylinder system with four sections.
For each section, the longitudinal displacement is written as a combination of sine and cosine functions. At the interfaces, displacement and force are continuous; at the gear end, the stress is zero; at the transducer end, the displacement is maximum and the strain is zero. Substituting these conditions gives a homogeneous system of equations for the section coefficients. A nontrivial solution exists only when the determinant of the coefficient matrix vanishes, which yields the overall frequency equation. The unknown conical-section length was found by solving this equation numerically.
The horn was made of 45 steel, with a density of 7890 kg/m³, Poisson’s ratio of 0.269, and Young’s modulus of 209 GPa. The cylindrical sections were 30 mm long, the large-end radius was 120 mm, and the small-end radius was 60 mm. The numerical solution placed the conical section at 130 mm for a nominal design frequency of 20 kHz. The geometric area ratio was 4, but the predicted amplitude gain at the gear face was lower because the gear mass and the groove geometry participate in the mode.
Finite-element analysis was used to check the design. The horn alone showed a longitudinal mode at 20,967 Hz. When the large gear was assembled to the horn, the mode nearest the design frequency, at 20,398 Hz, was torsional rather than longitudinal. This is the practical difficulty: a smooth horn and a large gear do not naturally produce a clean longitudinal mode. Helical grooves were therefore cut into the conical section. The grooves change the mass and stiffness distribution and introduce a controlled coupling between longitudinal and torsional motion. By adjusting the groove width and depth, the mode at 21,628 Hz was made predominantly longitudinal. The selected geometry was three uniformly spaced grooves with a 30° helix angle, a width of 16 mm, and a depth of 5 mm. An initial trial used a width of 15 mm and a depth of 4 mm; the wider and deeper groove gave the cleaner longitudinal mode.
| Parameter | Value |
|---|---|
| Material | 45 steel |
| Density | 7890 kg/m³ |
| Poisson’s ratio | 0.269 |
| Young’s modulus | 209 GPa |
| Large-end radius | 120 mm |
| Small-end radius | 60 mm |
| Cylindrical section length | 30 mm each |
| Conical section length | 130 mm |
| Groove number | 3 |
| Helix angle | 30° |
| Groove width | 16 mm |
| Groove depth | 5 mm |
| Design frequency | 20 kHz |
| Predicted resonance | 21,628 Hz |
| Node position | 93.04 mm from large end |
| Input amplitude | 5 µm |
| Output amplitude at gear face | 8.95 µm |
| Amplification | 1.79 |
The predicted resonance was about 8% above the nominal 20 kHz target, a shift that reflects the gear load and the grooves. The harmonic response placed the node at 93.04 mm from the large end. The maximum response occurred at 21,628 Hz, consistent with the modal result. In the transient analysis, a 5 µm sinusoidal displacement was applied at the large end. The particle at the gear face reached a maximum amplitude of 8.95 µm, a gain of 1.79. Its trajectory over one cycle was nearly a straight line, which confirms that the output was predominantly longitudinal. This amplitude was then used as the vibration input for the tooth-flank grinding simulation.
4. Representing the grinding wheel
The grinding wheel is modeled as a collection of rigid spherical grains. Scanning-electron microscopy of a 100-grit diamond grinding wheel shows irregular, angular grains with a broad size distribution. A spherical approximation is a compromise: it underestimates the sharpness of the cutting edges, but it preserves the size distribution and the kinematic footprint, and it keeps the computation tractable. The large negative rake angle of a grinding grain means that the contact is dominated by a rounded tip rather than a sharp wedge, which makes the spherical assumption less severe than it first appears.
The grain diameter is treated as a normal random variable. The mean is the average of the nominal maximum and minimum grit sizes, and the standard deviation is one third of the difference between the maximum and the mean. The maximum and minimum sizes are derived from the grit number. The grains are placed randomly on the grinding-wheel surface. A new grain is accepted only if its center is far enough from every existing grain:
$$\sqrt{(x_{gm}-x_{gn})^2+(y_{gm}-y_{gn})^2}\ge \frac{d_{gm}+d_{gn}}{2}$$
where \(d_{gm}\) and \(d_{gn}\) are the diameters of the two grains. If the condition is violated, the position is resampled. The volume fraction of grains is tied to the grinding-wheel structure number, so the same procedure can generate grinding wheels of different grit and structure. As an example, a grinding wheel with grit number \(M=60\) and structure number \(S=8\) was generated. The experimental grinding wheel was 120-grit alumina, and the same procedure was used with its grit number. The resulting surface has a random grain height and spacing, which is the input to the kinematic model.
The model does not resolve the bond bridges, the grain fracture, or the wear flats. Those features would require a measured grinding-wheel map rather than a statistical reconstruction. The statistical model is therefore best used for comparative studies—grinding wheel A versus grinding wheel B, or conventional versus ultrasonic grinding—rather than for absolute prediction of a specific grinding wheel.
5. Grain trajectories on the tooth flank
Four coordinate systems are used: the grinding-wheel frame, the grinding-wheel initial frame, the workpiece initial frame, and the workpiece moving frame. The workpiece rotates by \(\varphi_g=\omega_g t\) and feeds by \(L_g=v_g t\). The grinding wheel rotates by \(\varphi_w=\omega_w t\). The two axes are separated by the center distance \(E\) and inclined by the shaft angle \(\gamma\); the grinding wheel also has offsets \(a\), \(b\), and \(h\). The gear ratio is constant, \(i=\omega_g/\omega_w\). A homogeneous transformation \(\mathbf{M}_{gw}(t)\) maps a point from the grinding-wheel frame to the workpiece frame:
$$\mathbf{r}_g(t)=\mathbf{M}_{gw}(t)\,\mathbf{r}_{g0}$$
where \(\mathbf{r}_{g0}\) is the initial position of a point on a grain. For a spherical grain, the lowest point of the grain surface is the point that removes material. The trajectory of that point is the grain path.
In conventional grinding, the path is a trochoid on the curved flank. The grain enters the contact zone, reaches a maximum depth, and exits. The real-time depth \(a_s(t)\) changes continuously, and the engagement angle follows
$$\beta(t)=\arccos\left(1-\frac{2a_s(t)}{D_g}\right)$$
where \(D_g\) is the grain diameter. The cutting cross-section grows and then shrinks. Material displaced by the grain does not all become chip. A fraction is pushed to the sides and forms a raised ridge. The plowing model uses a cutting-efficiency ratio \(\xi=0.75\), so the pile-up area on one side is
$$A_p=\frac{1-\xi}{2}S$$
where \(S\) is the real-time cutting area. The ridge is approximated by a parabola; its height and width are set by the pile-up area. The extra height is added to the trajectory, so the final surface includes both the groove and the ridge.
In ultrasonic grinding, the gear is vibrated along its axis. The displacement is
$$u(t)=A_u\sin(2\pi f t+\phi_0)$$
where \(A_u\) is the amplitude, \(f\) is the frequency, and \(\phi_0\) is the initial phase. The transformation becomes
$$\mathbf{r}_u(t)=\mathbf{M}_{gw}(t)\left[\mathbf{r}_{g0}+\mathbf{e}_z A_u\sin(2\pi f t+\phi_0)\right]$$
where \(\mathbf{e}_z\) is the unit vector along the gear axis. The path is no longer a simple trochoid: it is a trochoid modulated by a high-frequency sinusoid. The grain can separate from the surface on the outward half of the cycle and re-enter on the inward half. The result is a sequence of overlapping, interrupted cuts. The groove is wider and shallower than in conventional grinding, and the ridges are broken into segments.
| Feature | Conventional grinding | Ultrasonic-assisted grinding |
|---|---|---|
| Contact mode | Continuous engagement | Intermittent engagement with periodic separation |
| Grain path | Trochoid on the tooth flank | Trochoid plus axial sinusoidal modulation |
| Depth history | Single rise-and-fall through the contact arc | Modulated rise-and-fall with repeated overlap |
| Groove pattern | Relatively continuous | Discontinuous, intersecting, widened |
| Ridge structure | Continuous side ridges | Broken and segmented ridges |
The trajectory equations include the real-time depth and the plowing ridge in both cases. The difference is the vibration term. This is important: the model does not treat ultrasonic grinding as a different material-removal law. It treats it as the same kinematic cutting process with an additional, high-frequency displacement. The change in topography comes from the changed path, not from an arbitrary roughness correction.
6. Building the surface by envelope computation
The tooth flank is discretized into a grid. Each grid point has an initial height. For each time step and each grain, the model computes the lowest point of the grain within a search region around the grid point. If that height is lower than the current height, the grid point is updated. The procedure is repeated for all grains and all time steps. The final surface is the lower envelope of all grain paths, which is equivalent to a Boolean minimum operation.
The regional approximation is a computational device. Testing every trajectory point against every grid point would be prohibitively expensive. Instead, for a grid point \((X_{P,Q},Y_{P,Q})\), a local search radius is defined, and the closest trajectory point is selected. The search radius is set by the expected grain footprint and the grid spacing. The error introduced by the local search is small compared with the uncertainty in the grain shape and the grinding-wheel topography.
The output is a height map \(z(x,y)\) on the flank. The areal roughness is
$$S_a=\frac{1}{A_m}\iint_{A_m} \left|z(x,y)\right|\,\mathrm{d}x\,\mathrm{d}y$$
where \(A_m\) is the measurement area and \(z\) is the height deviation from the reference plane. The areal parameter is used because the topography is three-dimensional and the groove direction changes across the flank. A profile parameter alone would miss the intersecting traces that are characteristic of ultrasonic grinding.
7. Predicted trends and experimental validation
The simulated flank shows a discontinuous groove pattern. The traces are not continuous lines; they are broken, overlapping, and crossed. This is the signature of intermittent contact. The conventional surface, by contrast, has more continuous grooves and a more regular ridge structure.
Two process variables were examined. At a fixed depth, increasing the grinding-wheel speed increases the number of active grains per unit time and reduces the uncut chip thickness per grain. The residual height distribution becomes more uniform, so the roughness falls. The conventional curve is relatively flat; the ultrasonic curve falls more strongly. The difference is attributed to the vibration, which reduces adhesion and grinding-wheel loading and prevents grains from concentrating their action in local patches. At a fixed speed, increasing the depth increases the uncut chip thickness, the plowing component, and the residual ridge height. Both conventional and ultrasonic roughness increase, but the ultrasonic case remains lower and rises more slowly. The vibration helps clear the chip and reduces the accumulation of displaced material.
The experimental check used a flat 45 steel specimen, 18 mm × 18 mm × 20 mm, ground with a 120-grit alumina grinding wheel on a surface grinder. The vibration frequency was 20 kHz, and the amplitude was varied from 6 µm to 18 µm. The depth was 0.01 mm, the grinding-wheel speed was 26 m/s, the feed was 4 m/min, and the process was dry. The surface was measured with a white-light interferometer. The conditions are listed in Table 5.
| Parameter | Value |
|---|---|
| Workpiece material | 45 steel |
| Workpiece size | 18 × 18 × 20 mm |
| Grinding wheel | 120-grit alumina |
| Ultrasonic frequency | 20 kHz |
| Ultrasonic amplitude | 6, 9, 12, 15, 18 µm |
| Grinding depth | 0.01 mm |
| Grinding-wheel speed | 26 m/s |
| Feed | 4 m/min |
| Environment | Dry |
| Measurement | White-light interferometry |
The measured surface showed the same discontinuous, intersecting groove pattern as the simulation. Both the simulated and the measured roughness decreased as the amplitude increased. The absolute values did not coincide. The difference is consistent with the simplifications in the model: machine vibration, grinding-wheel wear, material inhomogeneity, and measurement uncertainty are not included. The flat specimen is a special case of a curved surface; it isolates the trajectory and envelope algorithm from the gear geometry. It validates the kinematic and topographic part of the model, not the full gear-grinding operation.
The amplitude trend is not universal. One study on SiC ceramics found a minimum roughness at an amplitude of 0.216 µm, with roughness increasing at larger amplitudes. The difference is material-dependent: a brittle material can respond to impact by micro-fracture, while a ductile steel can respond by plastic flow. The model presented here is kinematic and does not include a material-damage criterion, so it should not be used to predict an optimum amplitude without calibration.
8. What the model captures and what it leaves out
The model captures the chain from nominal tooth geometry to acoustic output to grinding-wheel topography to grain kinematics to surface envelope. Its strengths are the explicit treatment of the curved flank, the stochastic grinding wheel, the real-time depth, the plowing ridge, and the ultrasonic modulation. It can explain why ultrasonic grinding produces discontinuous, intersecting grooves and why the roughness falls with grinding-wheel speed and amplitude, and rises with depth.
Several limitations remain. The gear–horn connection is treated as rigid; a clamped or adhesive interface would shift the resonance and the amplitude distribution. The horn was analyzed by finite-element analysis but not experimentally modal-tested. The gear-face amplitude was taken from a single point; the spatial distribution across the flank was not resolved. The grinding-wheel model uses spherical grains and a statistical size distribution; it does not include grain fracture, wear flats, or bond topology. The plowing model is phenomenological, with a fixed cutting-efficiency ratio. The material is treated as homogeneous, and the model does not couple temperature, residual stress, or subsurface damage. The full gear was not ground and measured; the experimental validation was performed on a flat specimen.
These limitations define the next steps. A measured grinding-wheel map would replace the statistical reconstruction. A thermo-mechanical model would connect the kinematic topography to residual stress and burn. A full gear experiment would test the geometric part of the model under production conditions. Experimental modal analysis of the horn–gear assembly would confirm the predicted resonance and node location. Within its scope, the model is useful for process design: it shows which parameters change the topography and why, and it provides a physical basis for selecting grinding-wheel speed, depth, and ultrasonic amplitude instead of relying on trial cuts.
The broader point is that surface topography in ultrasonic grinding is not an empirical add-on to conventional grinding. It is the result of a specific kinematic history on a specific surface. Once that history is represented explicitly, the effects of geometry, grinding-wheel topography, and vibration can be separated and compared. That is the practical value of the modeling chain for spiral bevel gears and for other complex surfaces finished by ultrasonic grinding.