Spiral bevel gears are used in right-angle drives where high torque, low noise, and long service life must coexist. In aerospace, electric-vehicle reducers, and precision machine tools, the final generating grinding operation after heat treatment is what restores tooth geometry and surface integrity. The process is sensitive because the wheel–flank contact is never constant. In double-flank generating grinding, the instantaneous contact line length and the undeformed chip volume both vary with the cradle angle, so the tangential grinding force fluctuates even when the nominal grinding parameters are fixed.
Those force fluctuations excite the wheel spindle, produce chatter marks on the flank, and raise the local contact temperature. At the peak, the temperature can exceed the phase-transition threshold of the case-hardened layer, causing burn and a loss of hardness. A constant generating speed cannot compensate for this geometry: it removes more material per unit time where the contact is aggressive and less where it is not. Sensor-based adaptive control can react, but it adds hardware, latency, and sensitivity to noise. The alternative is to precompute a speed schedule from the known tooth geometry so that the material removal rate remains nearly constant along the generating stroke.
Time-varying contact geometry and force fluctuation
For a fixed set of grinding conditions, the tangential force can be written as
$$
F_T=\frac{K_c}{\alpha+1}V\left(\frac{V_w}{V_s}\right)^{\alpha}\left(\frac{a_p}{D_e}\right)^{\alpha/2}
+\frac{K_f}{\alpha+1}l_c\left(\frac{V_w}{V_s}\right)^{\alpha}\left(\frac{a_p}{D_e}\right)^{\alpha/2}.
$$
Here \(F_T\) is the tangential grinding force, \(V\) is the instantaneous undeformed chip volume, \(l_c\) is the instantaneous contact line length, \(V_w\) is the workpiece generating speed, \(V_s\) is the wheel linear speed, \(a_p\) is the grinding depth, \(D_e\) is the equivalent wheel diameter, \(K_c\) and \(K_f\) are model coefficients, and \(\alpha\) is an exponent determined by the abrasive-edge distribution, typically between 0 and \(2/3\). The first term is the chip-deformation component; the second is the friction component.
In finish grinding, \(V_s\), \(a_p\), and \(D_e\) are held constant, and \(K_c\), \(K_f\), and \(\alpha\) are fixed by the wheel–workpiece pair. The force fluctuation therefore comes mainly from \(V\) and \(l_c\), both of which are functions of the cradle angle \(q_1\). The contact line is a geometric quantity. At a fixed cradle angle, the line is traced by changing the wheel rotation angle \(\theta\); its length is the sum of Euclidean distances between adjacent points on the traced curve. For double-flank grinding, the total contact line is the sum of the concave and convex flank contributions. For the gear in Table 1, the total contact line first rises and then falls, reaching a maximum near the middle of the 44° generating stroke. This shape is a direct geometric source of force fluctuation.
| Parameter | Value | Parameter | Value |
|---|---|---|---|
| Number of teeth | 33 | Radial cutter position | 90.1 mm |
| Face width | 38 mm | Cutter tilt angle | 0° |
| Spiral angle | 35° | Cutter swivel angle | 204.72° |
| Pitch cone angle | 64.13° | Vertical offset | 0 mm |
| Root cone angle | 59.98° | Blank mounting angle | 59.97° |
| Face cone angle | 66.45° | Horizontal offset | 0 mm |
| Addendum | 3.58 mm | Bed position | 0 mm |
| Dedendum | 8.69 mm | Roll ratio | 1.108401 |
| Working tooth height | 11 mm | Generating start angle | 81.72° |
| Crown distance | 48.77 mm | Center generating angle | 59.72° |
| Outer diameter | 217.63 mm | Generating end angle | 37.72° |
| Outer cone distance | 119.19 mm | Wheel diameter | 190.5 mm |
| Cutter point width | 3.3 mm | Cutter edge angle | 19.5° |
The chip volume is obtained from a CAD-based simulation of the generating motion. The model uses a rough-slotting–finish-grinding sequence, a 0.5° cradle step, and Boolean subtraction to isolate the undeformed chip. Representative values rise from 0.0007 mm³ at \(q_1=43.22^\circ\) to 0.0416 mm³ at \(q_1=48.72^\circ\), and the curve continues to change nonlinearly over the stroke. The important point is not the absolute value but the slope: the rate at which material is removed per unit cradle angle changes strongly, so a constant cradle speed produces a strongly varying removal rate.
Equal-volume segmentation and speed profile
Let \(V_{\rm rem}(q_1)\) be the cumulative volume removed and \(\omega_c=dq_1/dt\) the generating angular speed. The instantaneous removal rate is
$$
R(t)=\frac{dV_{\rm rem}}{dt}=\frac{dV_{\rm rem}}{dq_1}\omega_c.
$$
For constant \(R\), the generating angular speed must satisfy
$$
\omega_c(q_1)=\frac{R_0}{dV_{\rm rem}/dq_1},
$$
where \(R_0\) is the target removal rate. This is the control principle: slow down where the removed volume changes rapidly, speed up where it changes slowly. The practical implementation divides the stroke into \(N\) intervals and requires each interval to remove the same volume:
$$
\int_{q_1^{(i-1)}}^{q_1^{(i)}}\frac{dV_{\rm rem}}{dq_1}\,dq_1=\frac{V_{\rm total}}{N},
$$
where \(V_{\rm total}\) is the total volume removed in one tooth slot. When the removal rate is constant, each interval also takes the same time, \(T_0/N\), so the mean generating angular speed in interval \(i\) is
$$
\omega_c^{(i)}=\frac{\Delta q_1^{(i)}}{T_0/N}=\frac{N\Delta q_1^{(i)}}{T_0},
$$
where \(T_0\) is the total grinding time for one tooth slot and \(\Delta q_1^{(i)}=q_1^{(i)}-q_1^{(i-1)}\). The number of segments was set to \(N=7\). This is a compromise: more segments give a smoother force profile but increase the CNC program size and the load on the control system.
The same segmentation logic can be applied to different geometric objectives. Scheme 1 applies the equal-volume condition to the simulated chip-volume curve; scheme 2 applies the equal-integral idea to the contact-line curve. Table 2 lists the scheme 2 profile, expressed as the commanded generating speed in mm/min. The middle of the stroke runs at 209–309 mm/min, while the first and last segments run at 681 and 814 mm/min. This U-shaped profile redistributes time toward the region where the contact line is long and the force is highest.
| Segment | Start cradle angle (°) | End cradle angle (°) | Generating speed (mm/min) |
|---|---|---|---|
| 1 | 37.72 | 50.97 | 681 |
| 2 | 50.97 | 54.54 | 289 |
| 3 | 54.54 | 57.21 | 226 |
| 4 | 57.21 | 59.74 | 209 |
| 5 | 59.74 | 62.61 | 237 |
| 6 | 62.61 | 66.35 | 309 |
| 7 | 66.35 | 81.72 | 814 |
The contact-line profile performed better in finish grinding because the friction term is a large fraction of the total force. Contact line length is a direct geometric proxy for the wheel–flank contact area, so stabilizing it also stabilizes the frictional load and the thermal load in the contact zone. Equalizing chip volume mainly addresses the chip-deformation term; it is useful, but it does not control the friction-driven part of the fluctuation as directly.
Experimental validation
The tests were run on an H350GH CNC spiral bevel gear grinding machine. The workpiece material was 20CrMnTi; the wheel was a 32A80J8VS microcrystalline ceramic alumina wheel; the finish depth was 0.1 mm; the wheel speed was 20 m/s. Three schedules were compared: a constant generating speed of 900 mm/min, scheme 1, and scheme 2. Each schedule was repeated three times. Spindle current was recorded by the CNC system and used as an indirect indicator of grinding force through the current–force relationship; it is not a direct dynamometer measurement. Spindle vibration was measured with a PCB352C23 accelerometer. Tooth-surface error was measured on an L65G gear measuring center with a 25×25 point grid on each flank. The control itself used only the precomputed speed schedule; the current and vibration signals were used for evaluation, not for feedback.
Table 3 summarizes the measured response. The current values are three-run means; the vibration RMS values are from a representative run for each schedule.
| Metric | Constant 900 mm/min | Scheme 1: chip-volume profile | Scheme 2: contact-line profile | Scheme 2 reduction |
|---|---|---|---|---|
| Mean current fluctuation coefficient | 0.6924 | 0.3189 | 0.2820 | 59.27% |
| Spindle vibration RMS (m/s²) | 0.0481 | 0.0184 | 0.0175 | 63.62% |
| Tooth-surface error RMS (μm) | 1.276 | 1.071 | 0.849 | 33.46% |
The constant-speed runs showed the expected pattern: current rose and then fell, with a sharp peak and a strong fluctuation. The worst run reached a current fluctuation coefficient of 0.7446, and the three-run mean was 0.6924. Scheme 1 reduced the mean to 0.3189 (individual runs 0.3117, 0.3215, and 0.3236), a 53.94% reduction. Scheme 2 reduced it to 0.2820, a 59.27% reduction. The three scheme 2 runs gave 0.2895, 0.2508, and 0.3057, so the improvement was consistent across repeats.
The vibration data followed the same order. The constant-speed amplitude was about \(\pm0.13\) m/s² with an RMS of 0.0481 m/s². Scheme 1 reduced the amplitude to about \(\pm0.04\) m/s² and the RMS to 0.0184 m/s², a 61.75% reduction. Scheme 2 gave an RMS of 0.0175 m/s², a 63.62% reduction. The lower vibration is consistent with a smoother force trace and reduced excitation of the spindle.
Tooth-surface error also improved, though by a smaller relative amount. The constant-speed error RMS was 1.276 μm; scheme 1 reduced it to 1.071 μm (16.06%), and scheme 2 reduced it to 0.849 μm (33.46%). The smaller gain in geometric error than in force and vibration is not surprising: the final error also depends on wheel wear, machine thermal drift, fixture stiffness, and the accuracy of the generating motion itself. The speed schedule removes a major dynamic source of error, but it does not eliminate every static or quasi-static error source.
Process implications
The practical value of the method is that it acts at the motion level. It does not need a force dynamometer, a vibration sensor, or a fast feedback loop. The speed schedule is computed offline from the tooth-surface geometry, converted into a segmented CNC program, and executed synchronously with the cradle angle. This makes it easier to deploy on existing generating grinding machines than a closed-loop adaptive system, and it avoids the latency and noise sensitivity that come with sensor-based control.
The results also show a hierarchy among the geometric objectives. In finish grinding, where the wheel–workpiece contact is already fine and the friction term is significant, stabilizing the contact line is more effective than equalizing chip volume alone. The best schedule reduced the current fluctuation by 59.27%, the spindle vibration RMS by 63.62%, and the tooth-surface error RMS by 33.46% relative to the constant-speed baseline. These numbers are specific to the gear geometry, wheel, and machine used in the tests; a different gear or wheel specification requires a new segmentation calculation.
A natural next step is to make the segmentation adaptive. The present schedule is precomputed and fixed for a given workpiece. An online version could update the segment boundaries from spindle current, vibration, or acoustic emission, but that would reintroduce sensors and control latency. A more conservative route is to combine the feedforward schedule with periodic wheel-dressing compensation and thermal compensation, so that the geometric part of the error is handled by the speed profile and the slow drift is handled by the existing machine functions. That combination is likely to be more robust than either approach alone. For production, the immediate benefit is a lower-force, lower-vibration grinding cycle that can be programmed on the machine without additional hardware.