Spiral bevel gears are the default choice for intersecting-axis drives that must combine high torque, low noise, and long service life. Their curved teeth produce gradual engagement and a favorable contact pattern, but the same geometry makes final accuracy sensitive to heat-treatment distortion. Case carburizing to 58–60 HRC creates a wear-resistant surface and a tough core; the same cycle can shift the contact zone and degrade gear accuracy by two to three grades. Finishing is therefore not cosmetic. Grinding can correct distortion, but the equipment is expensive, stock removal is constrained by case depth, and the process is difficult to justify for small batches. Lapping removes less material and improves the contact pattern, yet near the pitch line the mating flanks have little relative sliding, so abrasive cutting is weak and removal tends to be nonuniform.
Ultrasonic-assisted lapping addresses that limitation by adding a high-frequency longitudinal vibration to one gear of the pair. A piezoelectric transducer generates the vibration, a horn amplifies it, and the gear transmits it to the lapping interface. The abrasive slurry experiences periodic impact and cavitation; intermittent contact can reduce friction and heat while improving slurry exchange. The acoustic chain, however, is not a rigid tool. The gear is both the working tool and the terminal load. Its mass, tooth geometry, and interface stiffness change the resonance of the coupled horn–gear system. A horn that resonates at 20 kHz in a free state may detune after the gear is mounted. The design problem is therefore best treated as a coupled waveguide, not as a horn with an arbitrary mass attached.
System Architecture and Non-Resonant Tuning
Non-resonant design keeps the gear geometry fixed and makes the horn adjustable. The assembled horn–gear unit is required to resonate at the generator frequency, typically 20 kHz. The horn is divided into three cylindrical sections and one conical section that represents the gear blank. The last cylindrical section, with axial length \(l_3\), is the tuning variable; the gear-equivalent section has length \(l_4\), equal to the tooth width. In a representative configuration, \(l_1=68\) mm, \(l_4=26\) mm, and \(l_3\) is allowed to vary from 0 to 50 mm.
For longitudinal motion, each uniform or slowly varying section is described by the one-dimensional wave equation
$$
\frac{d^2\xi_n}{dx^2}+\frac{1}{S_n}\frac{dS_n}{dx}\frac{d\xi_n}{dx}+k^2\xi_n=0,
\qquad k=\frac{\omega}{c},
\qquad c=\sqrt{\frac{E}{\rho}},
$$
where \(\xi_n(x)\) is the axial displacement, \(S_n(x)\) is the cross-sectional area, \(\omega\) is the angular frequency, \(E\) is Young’s modulus, and \(\rho\) is density. At each section interface, displacement and axial force are continuous. The transducer face is treated as a maximum-displacement, zero-strain boundary; the gear end is treated as a free boundary. Substitution into these conditions produces a homogeneous system whose coefficient determinant must vanish:
$$
\det\left[\mathbf{C}(f,l_3)\right]=0.
$$
For a prescribed frequency, this equation gives the required length of the tuning section. Conversely, for a fixed geometry, it gives the resonance frequency of the coupled system. The determinant is highly sensitive to the equivalent geometry used for the gear, which makes the load representation a primary design variable rather than a secondary detail.
Load Equivalence: Addendum, Pitch, or Root Circle
The gear can be replaced by an equivalent truncated cone in three ways: using the addendum circle, the pitch circle, or the root circle as the reference radius. All three keep the tooth width \(l_4=26\) mm but differ in equivalent radius \(R_4\) and cone angle. The geometric differences are substantial enough to change the tuned horn length and the predicted resonance frequency.
| Equivalent model | \(R_4\) (mm) | Cone angle | Tuned \(l_3\) (mm) | Theoretical frequency deviation from 20 kHz |
|---|---|---|---|---|
| Addendum circle | 21.66 | 14.77461° | 30.96 | 8.877% |
| Pitch circle | 16.648 | 11.04094° | 34.88 | 5.353% |
| Root circle | 13.94 | 9.387259° | 47.32 | 8.612% |
The pitch-circle and root-circle models differ by 12.44 mm in tuned horn length, and the addendum model is 16.36 mm shorter than the root model. Such a spread is not a minor modeling uncertainty. It changes the physical length of the horn, the position of the displacement node, and the location available for the mounting flange. The determinant zero is also unique within the practical \(l_3\) range of 10–50 mm for all three models, which means that each equivalence rule leads to a self-consistent but different physical design.
Simulation and Experimental Validation of Simplified Loads
Modal and harmonic analyses were used to compare the three equivalent systems with a full toothed spiral bevel gear model. The modal search covered 15–25 kHz and extracted the first ten modes; the harmonic analysis applied a 5 μm axial displacement at the input and scanned 40 frequency steps of 100 Hz around the longitudinal mode. The full toothed model resonated at 20.306 kHz and produced an axial output of 9.3 μm.
| Model | Simulated \(f\) (kHz) | Simulated \(\xi_o\) (μm) | Measured \(f\) (kHz) | Measured \(\xi_o\) (μm) | Frequency difference from simulation | Amplitude difference from simulation |
|---|---|---|---|---|---|---|
| Addendum circle | 19.005 | 12.1 | 18.154 | 13.312 | 4.58% | 9.539% |
| Pitch circle | 21.645 | 7.7 | 21.433 | 6.827 | 0.984% | 12.019% |
| Root circle | 23.169 | 4.7 | 23.894 | 5.628 | 3.081% | 17.971% |
| Full toothed gear | 20.306 | 9.3 | 21.259 | 7.964 | — | — |
The pitch-circle equivalent gave the closest overall agreement. Its measured impedance frequency differed from the full gear by 0.815%, and its measured axial amplitude differed by 15.374%. The addendum and root models produced larger mismatches: 15.756% and 11.671% in frequency, and 50.273% and 34.373% in amplitude, respectively. The addendum model overestimated the output amplitude because it represented a larger effective mass and stiffness distribution near the tooth tip; the root model underestimated the amplitude and shifted the resonance upward. The pitch-circle equivalence was therefore adopted for subsequent horn design.
Horn Material and Damped Flange Interface
Material selection is constrained by acoustic impedance, fatigue strength, machinability, and available space inside the lapping machine spindle. Three horn materials were compared while keeping the gear as 20CrMnTi carburized steel. The relevant properties are listed below; nylon 66 is included because it is used as an interface damping element rather than as a horn material.
| Component | Material | \(E\) (GPa) | \(\rho\) (kg/m³) | Poisson ratio |
|---|---|---|---|---|
| Gear | 20CrMnTi carburized steel | 207 | 7800 | 0.25 |
| Horn | TC4 titanium alloy | 111 | 4429 | 0.338 |
| Horn | 45 steel | 209 | 7850 | 0.269 |
| Horn | Aluminum alloy | 68 | 2840 | 0.33 |
| Damping washer | Nylon 66 | 3.5 | 1150 | 0.39 |
At 20 kHz, the frequency equation required \(l_3=35.18\) mm for 45 steel, 43.86 mm for TC4, and 41.8 mm for aluminum alloy. The corresponding frequency roots were 21.9 kHz, 21.4 kHz, and 19.28 kHz. The 45 steel horn therefore produced the shortest tuning section, 6.68 mm shorter than TC4 and 8.68 mm shorter than aluminum. This compactness matters because the horn must fit within the machine spindle envelope. The 45 steel horn also has better mechanical compatibility with the 20CrMnTi gear than the lower-modulus aluminum alloy, so the tuning length was rounded to 35 mm for manufacture.
The mounting flange must be placed near a displacement node to avoid coupling the ultrasonic system to the machine frame. Harmonic analysis of the full gear system placed the node at 65.167 mm from the input face. A double-flange module was used at this position: a 45 steel flange body for structural rigidity and nylon 66 gaskets at the flange–horn and flange–spindle interfaces. The resulting mechanical–damping connection is intended to preserve axial energy transfer while suppressing interface heating, plastic deformation, and parasite vibration from the machine tool.
Modal analysis of the complete system, including the flanges and gaskets, gave a longitudinal resonance of 21.964 kHz. Impedance testing measured 21.259 kHz, a difference of 3.262%. The simulated axial output was 8.3 μm; laser Doppler vibrometry measured 7.964 μm, a difference of 4.132%. These deviations are consistent with the compliance of the threaded and gasketed interfaces, which cannot be represented as perfectly bonded contacts in the finite-element model.
Sensitivity to Tooth Width and Helix Angle
In production, tooth width \(b\) and helix angle \(\beta\) vary within manufacturing tolerances and change the acoustic load. Tooth width alters the axial length and stiffness of the gear section; helix angle alters the contact geometry and the direction in which the longitudinal wave is decomposed at the tooth surface. To quantify these effects, gear models were generated with \(b=24\)–28 mm in 0.5 mm increments and with \(\beta=31\)–39° in 1° increments. Modal–harmonic simulations were then compared with impedance and vibration measurements on machined samples.
| Parameter | Simulated frequency trend | Simulated axial amplitude trend | Measured frequency trend | Measured axial amplitude | Measured tooth-surface amplitude |
|---|---|---|---|---|---|
| Tooth width \(b\), 24–28 mm | Decreases by about 51–119 Hz per 0.5 mm | Decreases with increasing \(b\), roughly 0.6–2.8 μm per 0.5 mm | Decreases by about 33–122 Hz per 1 mm | 14.942 μm at \(b=24\) mm to 4.315 μm at \(b=28\) mm; step decrease 1.312–4.402 μm/mm | 0.786–1.428 μm |
| Helix angle \(\beta\), 31–39° | 20.115–21.395 kHz, overall downward; about 10–80 Hz per degree | Decreases by roughly 0.05–0.41 μm per degree | 21.344 kHz at 31° to 21.226 kHz at 39°; about 11–60 Hz per 2° | 7.153–8.362 μm, nonmonotonic | 0.862–1.362 μm |
Tooth width is the stronger parameter. Increasing \(b\) lowers the resonance frequency and the axial output, but the response is asymmetric: reducing \(b\) below the nominal 26 mm raises the amplitude more than increasing \(b\) above 26 mm lowers it. An undersized tooth width is therefore more damaging to amplitude stability than an oversized one. The measured axial amplitude fell from 14.942 μm at \(b=24\) mm to 4.315 μm at \(b=28\) mm, while the tooth-surface amplitude remained in the range 0.786–1.428 μm and fluctuated rather than following a smooth trend.
Helix angle has a weaker influence on the global resonance but a clear influence on the tooth-surface vibration components. The measured axial amplitude varied between 7.153 and 8.362 μm without a monotonic trend, whereas the concave and convex flank amplitudes generally decreased as \(\beta\) increased. The local tooth-surface amplitudes remained between roughly 0.9 and 1.4 μm. Such fluctuation can produce local overcutting during ultrasonic lapping, but controlled fluctuation may also promote micro-scale refinement of the flank. The net effect on material removal and surface integrity is not resolved by frequency and amplitude measurements alone.
Preload Torque of the Threaded Gear–Horn Joint
The gear is joined to the horn through a threaded shaft. Low preload allows micro-slip and loss of self-locking under vibration; excessive preload raises thread-root stress and accelerates fatigue. The joint was therefore tested at preload torques from 20 to 45 N·m in 5 N·m increments. Impedance analysis gave the resonance frequency and mechanical quality factor \(Q_m\), while laser vibrometry gave the axial output amplitude.
| Preload torque (N·m) | Resonance frequency \(f\) (kHz) | Mechanical quality factor \(Q_m\) | Axial output amplitude (μm) |
|---|---|---|---|
| 20 | 21.336 | 165.357 | 7.643 |
| 25 | 21.434 | 253.456 | 7.752 |
| 30 | 21.412 | 344.622 | 7.913 |
| 35 | 21.508 | 462.679 | 7.994 |
| 40 | 21.574 | 471.05 | 7.991 |
| 45 | 21.596 | 480.874 | 8.018 |
Both resonance frequency and \(Q_m\) increase with preload, but the gain flattens above 35 N·m. The axial amplitude follows the same pattern: it rises from 7.643 μm at 20 N·m to 7.994 μm at 35 N·m and then changes only marginally to 8.018 μm at 45 N·m. A preload of 35 N·m was therefore selected as the operating value. It provides a stable contact stiffness and a high quality factor without driving the thread root into the high-stress regime associated with the upper end of the torque range.
Design Implications and Limits
The results support a coherent design sequence for a 20 kHz spiral bevel gear ultrasonic lapping system. First, represent the gear with a pitch-circle equivalent cone rather than an addendum or root approximation; the pitch-circle model gives the smallest mismatch in both resonance frequency and axial amplitude. Second, use 45 steel for the horn when spindle space is limited and the gear is 20CrMnTi steel; the shorter tuning length and closer mechanical match reduce the penalty of the coupled load. Third, place a double flange at the displacement node and interpose nylon gaskets to create a mechanical–damping interface; this preserves axial transmission while reducing coupling to the machine frame. Fourth, control the threaded joint at approximately 35 N·m, where \(Q_m\) and amplitude have reached a plateau. Fifth, treat tooth width as a tighter acoustic tolerance than helix angle: a 1 mm change in \(b\) shifts the resonance by tens to more than one hundred hertz and changes the axial amplitude by several micrometres, whereas a 2° change in \(\beta\) has a smaller global effect but still alters the local flank vibration.
The remaining uncertainty lies in the transfer from acoustic response to lapping performance. The simplified one-dimensional waveguide model neglects shear, rotation, and the detailed contact stiffness of the tooth flank; the finite-element model cannot fully reproduce the gasketed and threaded interfaces; and the measured tooth-surface amplitudes fluctuate in a range that is small in absolute terms but large relative to the local vibration component. These effects explain why measured resonance frequencies and amplitudes differ from simulation by several percent, and why the flank amplitude should be monitored rather than inferred from the axial output alone. For production use, frequency tracking and preload verification are more reliable than a one-time room-temperature tuning step, because the acoustic load changes with tooth geometry, contact state, and thermal condition during actual lapping.