Spiral bevel gears transmit motion between intersecting shafts through curved teeth and conical pitch surfaces. In aerospace, helicopter, and other high-power drivetrains, the tooth flank must carry high loads while maintaining contact fatigue resistance, low noise, and stable transmission efficiency. Surface roughness and microtopography are not secondary details in this context: they influence film formation, contact pressure distribution, running-in behavior, and the initiation of surface damage. Ultrasonic-assisted grinding introduces a high-frequency, low-amplitude vibration into the grinding contact. The abrasive grains no longer follow purely continuous paths; the contact becomes intermittent, the trajectories overlap differently, and the resulting surface can have shallower residual ridges and a more uniform bearing profile. Predicting that surface on a spiral bevel gear is difficult because the flank is a complex spatial surface and the grinding kinematics vary across the tooth width and tooth height.
The surface-generation problem
In conventional grinding, a grit engaged with the workpiece removes material along a continuous path. The final topography is the lower envelope of many overlapping scratches, with ridges left between adjacent passes. In ultrasonic grinding, an axial vibration term is superimposed on the relative motion, so the grit path oscillates at the ultrasonic frequency while the wheel and workpiece continue their normal motions. The result is not simply a deeper or shallower scratch. The path becomes wavy, adjacent passes cross at different phases, and the contact alternates between engagement and separation. This changes the distribution of residual height and can produce discontinuous grooves rather than continuous scratch fields.
A predictive model therefore needs four linked elements: the spatial tooth-surface geometry, the vibration state at the gear end, a statistical description of the abrasive wheel, and the kinematic trajectory of individual grits including real-time depth and plowing. The approach is essentially kinematic–geometric. It predicts the envelope left by abrasive trajectories and uses empirical treatment for the plowing ridge; it is not a full thermo-mechanical simulation of chip formation.
Tooth-surface geometry as the base frame
The gear model is built from the generating motion of a cutter disk. The cutter surface is defined in the cutter-disk coordinate system, and a sequence of homogeneous transformations maps it into the gear coordinate system. The contact condition follows the local conjugate principle: at the instantaneous contact point, the two surfaces share a common normal and satisfy the meshing equation. For the large wheel, the cutter radius, cutter point width, radial cutter position, angular cutter position, and machine settings determine the generated flank. The pinion surface is obtained in the same way but with its own generating parameters and modified roll terms.
For the example gear pair, the tooth surface was discretized by projecting the flank onto an axial section, solving the boundary curves, and meshing the projected region. A 5 × 9 grid was used for the large-wheel example. The resulting three-dimensional points were imported into a solid modeler, where lofted surfaces, stitching, and circular arrays produced the complete tooth form. This geometric model supplies the spatial domain over which the grinding trajectories are evaluated.
| Parameter | Value |
|---|---|
| Shaft angle | 90° |
| Module | 3.6 mm |
| Number of teeth, pinion / wheel | 41 / 43 |
| Spiral angle | 25° |
| Pressure angle | 20° |
| Face width | 30 mm |
| Addendum coefficient | 0.887 |
| Clearance coefficient | 0.677 |
| Addendum / dedendum | 3.196 mm / 3.6 mm |
| Hand, large wheel | Right |
Acoustic design for longitudinal vibration at the gear end
A spiral bevel gear is a large, non-resonant load compared with a conventional ultrasonic horn. Treating the whole assembly as a set of independently resonant sections is therefore not sufficient. The design used a non-resonant approach: the gear was simplified as a truncated cone, and the horn–gear assembly was treated as a composite longitudinal vibration system. The horn consists of two cylindrical sections, a conical transition, and the gear body. Continuity of displacement and force at each interface yields a set of linear equations for the section amplitudes. A non-trivial solution requires the coefficient determinant to vanish:
$$ \det[\mathbf{D}(f,l_2)]=0 $$
where \(f\) is the design frequency and \(l_2\) is the unknown conical transition length. With the cylindrical sections fixed at 30 mm each and the large- and small-end radii at 120 mm and 60 mm, respectively, the frequency-equation error curve crossed zero at \(l_2=130\) mm.
Finite element modal analysis of the plain composite horn gave a longitudinal mode at 20,967 Hz. When the gear was assembled to the horn, however, the mode nearest the design frequency was torsional rather than longitudinal; a mode at 20,398 Hz showed a twisted displacement pattern and did not meet the requirement. Helical grooves were therefore introduced into the conical section to redistribute mass and stiffness and to modify the mode shape. Among the tested groove geometries, three helical grooves with a 30° rotation angle, 16 mm width, and 5 mm depth produced a predominantly longitudinal mode at 21,628 Hz.
Harmonic response analysis placed the axial node at 93.04 mm from the large end. Transient analysis applied a 5 µm axial displacement at 21,628 Hz. A particle on the gear end face traced a nearly straight path, with a maximum axial amplitude of 8.95 µm, corresponding to a gain of 1.79. That output amplitude and frequency become the vibration input for the grinding-trajectory model. The acoustic design was assessed numerically; experimental validation of the full horn–gear assembly is a separate step.
| Acoustic parameter | Value |
|---|---|
| Horn material | 45# steel |
| Density | 7890 kg/m³ |
| Poisson ratio | 0.269 |
| Elastic modulus | 209 GPa |
| Large-end radius | 120 mm |
| Small-end radius | 60 mm |
| Cylindrical section lengths | 30 mm each |
| Conical transition length | 130 mm |
| Helical grooves | 3, 30° rotation, 16 mm wide, 5 mm deep |
| Longitudinal resonance | 21,628 Hz |
| Node location | 93.04 mm from large end |
| Input / output amplitude | 5 µm / 8.95 µm |
| Amplitude gain | 1.79 |
The abrasive wheel as a stochastic surface
Scanning electron microscopy of a 100-grit diamond wheel showed irregular grain shapes, variable protrusion heights, and non-uniform spacing. Modeling every irregular facet would multiply the computational cost without necessarily improving the statistical envelope. Because the effective rake angle of a grinding grit is strongly negative, a spherical approximation is a practical compromise. Each grain is represented by a sphere whose diameter is drawn from a normal distribution, and the grains are placed randomly on the wheel surface subject to a non-interference condition:
$$ \sqrt{(x_{gm}-x_{gn})^2+(y_{gm}-y_{gn})^2}\ge \frac{d_{gm}+d_{gn}}{2} $$
where \(x_{gm},y_{gm}\) and \(d_{gm}\) are the position and diameter of grain \(m\), and the subscript \(n\) denotes a different grain. If a newly placed grain violates this condition, its position is reassigned until the constraint is satisfied. A uniform-distribution model was also generated as a baseline. The simulation example used an alumina wheel with grit number \(M=60\) and structure number \(S=8\); the experimental validation used a 120-grit alumina wheel.
Kinematics and the ultrasonic trajectory
The grinding model uses four coordinate frames: the wheel moving frame, the wheel initial frame, the workpiece initial frame, and the workpiece moving frame. The composite homogeneous transform includes the workpiece rotation \(\phi_g=\omega_g t\), the feed \(L_g=v_g t\), the center distance \(E\), the shaft angle \(\gamma\), and the wheel rotation \(\phi_w=\omega_w t\). For a grit at position \(\mathbf{r}_s\) in the wheel frame, the path in the gear frame is
$$ \mathbf{r}_g(t)=\mathbf{M}_{gw}(t)\,\mathbf{r}_s $$
where \(\mathbf{M}_{gw}\) is the composite transform from the wheel to the gear. Each grit is assigned a phase angle and a radial offset so that grits enter the cutting zone in sequence rather than as a single uniform front.
In ultrasonic grinding, an axial vibration term is superimposed in the machine frame:
$$ u_z(t)=A\sin(2\pi f t+\lambda_0) $$
where \(A\) is the amplitude, \(f\) is the ultrasonic frequency, and \(\lambda_0\) is the initial phase. The instantaneous cutting depth is not constant. As a spherical grit enters and leaves the engagement zone, the effective depth follows the local geometry of the uncut surface. The model writes this as
$$ a_s(t)=a_e-\Delta z_s(t) $$
where \(a_e\) is the set depth and \(\Delta z_s\) is the instantaneous gap between the grit surface and the uncut surface. The trajectory is also modified by plowing. Grinding contact passes through sliding, plowing, and cutting stages. In the plowing stage, material is displaced to the sides of the scratch and forms ridges instead of being fully removed as a chip. The ridge area is related to the instantaneous cutting area \(S\) by
$$ A_p=\frac{1-\xi}{2}S,\qquad \xi=0.75 $$
where \(\xi\) is the cutting-efficiency ratio. The ridge profile is approximated as a parabola, and the height and width of the parabola are updated with the real-time depth. This gives a lower envelope with side ridges rather than a simple series of semicircular scratches.
From trajectories to topography and roughness
The final surface is generated by a region-approximation algorithm. The tooth surface is first discretized into a height field. For each grid point, the algorithm searches the neighboring region for the lowest contact height produced by any grit trajectory. If that height is lower than the current surface height, the grid value is replaced. The procedure is repeated for all grid points and all grit passes. The resulting height field is displayed as a three-dimensional surface, and the roughness is evaluated as
$$ S_a=\frac{1}{A}\iint_A |z(x,y)|\,dx\,dy $$
where \(A\) is the measurement area and \(z(x,y)\) is the height deviation from the reference plane. In the ultrasonic case, the simulated surface shows discontinuous grooves and crossing trajectories. In conventional grinding, the scratches are more continuous and the residual ridges are more regular. That difference is the geometric signature of the intermittent contact.
Parametric trends
Simulations were run for ordinary and ultrasonic grinding over a range of wheel speeds and grinding depths. The general trends are summarized in the table below. Increasing wheel speed raises the number of active grits per unit time and makes the coverage more uniform. In conventional grinding, the roughness decreases only slightly because the residual height distribution becomes more even but remains dominated by the deepest scratches. In ultrasonic grinding, the decrease is stronger: the vibration reduces adhesion and loading of the wheel surface, discourages grit concentration in local zones, and spreads the cutting action over a wider area.
Increasing grinding depth raises the undeformed chip thickness, strengthens plowing, and increases the residual ridge height. Both conventional and ultrasonic roughness therefore increase with depth. The ultrasonic case still gives a lower roughness at the same depth because the vibration reduces the negative effects of force and temperature, limits adhesion, and assists chip clearance. The trend is a mitigation, not an elimination, of depth-induced surface deterioration.
| Variable | Conventional grinding | Ultrasonic grinding |
|---|---|---|
| Wheel speed increases | Roughness decreases slightly | Roughness decreases more strongly; less adhesion and local loading |
| Grinding depth increases | Roughness increases | Roughness increases, but remains lower than conventional at the same depth |
Experimental check with a planar workpiece
Direct ultrasonic grinding of a complete spiral bevel gear is experimentally demanding. The trajectory formulation was therefore reduced to the special case of a planar workpiece, which has zero curvature, and a planar grinding test was used to check the surface-generation method. The workpiece was 45# steel with dimensions 18 × 18 × 20 mm. The machine was an M7130 surface grinder, the wheel was 120-grit alumina, and the ultrasonic vibration was applied at 20 kHz. The amplitude was varied over 6, 9, 12, 15, and 18 µm. The grinding depth was 0.01 mm, the wheel speed was 26 m/s, the feed rate was 4 m/min, and the process was run dry. Surface topography was measured with an NT9100 white-light interferometer.
| Grinding or validation parameter | Value |
|---|---|
| Wheel model in simulation | Alumina, grit 60, structure 8 |
| Experimental wheel | Alumina, grit 120 |
| Workpiece for validation | 45# steel, 18 × 18 × 20 mm |
| Ultrasonic frequency in validation | 20 kHz |
| Ultrasonic amplitude in validation | 6, 9, 12, 15, 18 µm |
| Grinding depth | 0.01 mm |
| Wheel speed | 26 m/s |
| Feed rate | 4 m/min |
| Environment | Dry grinding |
| Surface measurement | White-light interferometer, NT9100 |
Both the simulated and measured roughness decreased as the ultrasonic amplitude increased from 6 µm to 18 µm. The measured surface also showed the characteristic morphology of ultrasonic grinding: grooves were not continuous across the field but appeared as interrupted segments, and neighboring trajectories crossed and interfered. The simulation reproduced this pattern, which supports the trajectory-based generation method. The agreement was not exact. Differences were attributed to simplifications in the geometric model, machine vibration, wheel wear, and material effects that were not included in the simulation. The planar test is an indirect validation of the spiral bevel gear model, not a full gear-grinding experiment.
What the model captures and what it leaves out
The model captures the geometric chain from tooth-surface generation to abrasive trajectory to roughness. It shows why ultrasonic grinding changes the surface not only by reducing the mean depth of cut but by changing the overlap and phase of grit paths. It also provides a way to connect acoustic design variables, such as resonant frequency, node position, and output amplitude, to a measurable surface parameter. For process development, that connection is more useful than a single roughness number: it allows wheel speed, depth, vibration amplitude, and groove geometry to be compared within the same kinematic framework.
The limitations are equally important. The acoustic assembly was evaluated by finite element modal, harmonic, and transient analyses, but the full horn–gear system was not experimentally validated in the reported design. The grinding model treats the wheel as a set of spherical grits with a statistical size distribution and does not include wheel wear, grit fracture, machine-tool vibration, thermal softening, or microstructure-level material behavior. The experimental check used a planar 45# steel workpiece, so the curvature and contact conditions of an actual spiral bevel gear were not reproduced in the test. The model is therefore best used as a design and trend-prediction tool, not as a final certification of tooth-flank roughness. Direct gear-grinding tests and wear-dependent wheel models would be needed before the predicted roughness could be used as a production acceptance criterion.
For spiral bevel gear manufacturing, the practical value of the approach lies in the coupling it exposes. Tooth geometry determines where the grit paths act; the acoustic system determines the amplitude and phase of the vibration; the wheel statistics determine the distribution of cutting edges; and the kinematic model determines how those edges overlap. When these factors are treated separately, ultrasonic grinding can appear as a simple roughness-reduction step. When they are coupled, the surface is seen as the result of a specific trajectory envelope, and the process parameters can be selected to control that envelope rather than adjusted by trial alone.