A spiral bevel gear carries the full torque path of a helicopter main reducer: it converts engine output into rotor-shaft power while its teeth absorb the alternating loads exchanged between rotor head and airframe. Load capacity, service life, dry-running capability and stable operation at elevated temperature, pressure and shaft speed all depend on the tooth flank geometry. Chamfering the tooth edges is a small geometric operation with disproportionate consequences — it removes the sharp feather edge running along the tooth length, lowers the impact and noise generated at each mesh cycle, prevents edge chipping and reduces stress concentration at the flank boundary. On a case-hardened spiral bevel gear with curved tooth traces, however, that edge is both hard and three-dimensionally curved, which is exactly the combination that defeats manual filing and general-purpose chamfering machines.
The automation problem
Manual chamfering of hardened spiral bevel gears is skilled work: the operator must follow a helical edge whose curvature changes continuously from heel to toe, on a surface at roughly HRA 82, while removing a controlled amount of material. The result is labour-intensive, sensitive to operator technique and therefore weakly repeatable from tooth to tooth and from part to part. Dedicated chamfering machines remove the operator but tie the process to one part family and one setup, with high capital cost and limited tolerance for geometry changes.
A robot-based approach trades dedicated hardware for a general six-axis kinematic platform plus a compliant tool. Two requirements dominate the design: the tool must reach every tooth edge without collision or singularity, and the contact force must be regulated closely enough that the chamfer width stays inside tolerance on a curved, hardened edge where depth of cut cannot be set mechanically by a fixture.
System architecture
The cell combines a six-axis industrial robot, a floating pneumatic spindle, an indexing rotary table and the workpiece. The robot is an ABB IRB4600 with a working range of 2.05 m and a maximum payload of 60 kg. The spindle is an Amtru FB-CUT110 pneumatic unit with a radial floating function: its compliance lets the tool follow local changes in profile curvature instead of forcing a rigid path onto the part. Rated floating force reaches 25 N and the maximum spindle speed is 19,000 r/min. A diamond grinding disc held in a special tool holder is mounted on the spindle; the gear is clamped on the rotary table.
The kinematic division of labour matters. The robot positions and orients the spindle along the tooth edge; the table indexes the gear by a programmed angle between teeth. Because indexing is decoupled from the robot path, a single calibrated trajectory for one tooth can be repeated around the whole gear, which is what makes the process economical on a 31-tooth part.
Process chain: from frames to finished tooth
The workflow runs through tool and workpiece frame calibration, offline path programming, automatic grinding and dimensional inspection.
Tool centre point calibration
The tool centre point (TCP) is established with the robot’s built-in four-point routine, in which four taught joint targets define points on a sphere whose centre is the TCP in the world frame. The underlying pose relation is
$$B = AX \qquad \Rightarrow \qquad X = A^{-1}B$$
where A is the current position expressed in tool0, B the same position expressed in the active tool frame and X the tool pose being solved for. Because a disc-shaped grinding wheel cannot be brought to four distinct contact points reliably, a pointed probe of identical length is substituted during teaching; the resulting TCP corresponds to the centre of the disc end face, which is also the point used as the grinding reference during path generation.
Workpiece frame and path generation
The workpiece frame is set by a three-point method, after which the gear’s 3D model is imported into ABB Robot Studio and translated so that model coordinates coincide with the measured part frame. The chamfer path is then generated directly in the workpiece frame and post-processed into robot targets, each carrying position and orientation.
Two path decisions come from shop-floor comparison rather than from geometry alone. First, the disc rotates clockwise. Second, the tool travels from the tooth root outward toward the tooth tip; this direction produced better edge quality than the reverse, tip-to-root direction. Before download, the trajectory is simulated to keep the wrist away from singular configurations and to keep the spindle and tool holder clear of the adjacent teeth and fixture, which is the tightest constraint on a bevel gear where neighbouring flanks obstruct the approach.
Floating compensation and the grinding force
The floating spindle converts a position-controlled robot into a force-compliant finishing tool. Radial compliance accommodates curvature variation along the edge, absorbs part-to-part scatter and small deviations between the actual and nominal profile, and limits the depth of cut when the path and the real edge diverge — the mechanism that prevents the disc from cutting into the gear body. The compensation and retraction stroke reaches ±5 mm. A pressure-control circuit sets the contact force, so the elastic deformation of the floating element, and hence the contact pressure at the disc edge, is adjustable through the supply pressure.
Because the disc contacts the tooth edge with its periphery, the cut approximates vertical surface grinding. The resultant force is resolved into a normal component Fn, a tangential component Ft and an axial component Fa; the same decomposition applies at the level of a single active grain, giving F′n, F′t and F′a. With the specific grinding force σ (force per unit chip cross-section measured normal to the cutting direction) and the mean chip area am,
$$F’_t = \sigma a_m$$
$$F’_n = \lambda F’_t = \lambda \sigma a_m$$
where λ is the ratio of normal to tangential grain force. Summing over the j cutting edges engaged simultaneously across contact width b and contact arc length lc yields the macroscopic components:
$$F_t = jF’_t = \sigma b a_p \frac{v_w}{v_s}$$
$$F_n = jF’_n = \lambda \sigma b a_p \frac{v_w}{v_s}$$
$$F_a = jF’_a = \sigma b a_p \frac{v_w v_a}{v_s^{2}}$$
Here ap is the axial infeed of the disc, vw the workpiece speed at the contact, vs the peripheral speed of the disc at the active point and va the axial component of vs. The equations explain the two control levers available to the operator. Raising the floating pressure increases ap, so more material is removed per unit time, each grain cuts a thicker chip and all three force components rise. Raising spindle speed at a fixed feed reduces the number of grains passing per unit time in proportion, thins the chip and lowers the force per grain — the usual route to better surface finish without increasing the load on the compliant tool.
Validation on a helicopter reducer gear
The cell was evaluated on a spiral bevel gear from a helicopter main reducer, supplied as a production part in alloy steel with a chamfer-zone hardness of HRA 82. The specification called for a chamfer face of C0.5–C0.75 mm. Following the process chain above, the tool and workpiece frames were calibrated, the path was generated offline in Robot Studio, singularities and interference zones were eliminated in simulation, and the trajectory was downloaded. After each tooth the table indexed to the next position and repeated the same cycle.
Chamfer width was measured with a Model 334006 contact profilometer. To check consistency both around the gear and across the tooth length, 8 of the 31 teeth were selected at random and each was measured at three locations — top, middle and bottom of the flank.
| Tooth No. | Top | Middle | Bottom | Mean |
|---|---|---|---|---|
| 1 | 0.722 | 0.723 | 0.723 | 0.722 6 |
| 5 | 0.722 | 0.725 | 0.722 | 0.723 0 |
| 9 | 0.724 | 0.722 | 0.722 | 0.722 6 |
| 13 | 0.723 | 0.726 | 0.726 | 0.725 0 |
| 17 | 0.722 | 0.725 | 0.721 | 0.722 6 |
| 21 | 0.722 | 0.722 | 0.723 | 0.722 3 |
| 25 | 0.722 | 0.721 | 0.726 | 0.723 0 |
| 29 | 0.723 | 0.720 | 0.722 | 0.721 6 |
All 24 readings fall within a 6 µm band, from 0.720 mm to 0.726 mm. The mean tooth values span 0.7216–0.7250 mm, a range of 3.4 µm, and the overall mean is 0.7228 mm. Within any single tooth the three positions differ by no more than 6 µm, so the chamfer neither tapers along the flank nor drifts as the table indexes around the gear. Consistency of this order is beyond what manual chamfering can hold on a hardened curved edge, and it is obtained without operator intervention during the cycle.
The absolute level is also informative: at 0.72 mm nominal, the process sits near the upper bound of the C0.5–0.75 mm specification window. The path offset that fixes the target width is a programming parameter, so the same trajectory can be biased toward the centre of the tolerance band if a different nominal is required, without changing the hardware or the force setting.
Engineering implications and operating envelope
Three characteristics define where this configuration fits. First, it is a finishing and edge-preparation process, not a stock-removal operation: with a floating force capped at 25 N and a compliant stroke of ±5 mm, the material removed per pass is small by design, and the value lies in controlled edge geometry rather than in throughput. Second, the compliance window sets the permissible mismatch between programmed path and actual part — fixture error, clamping repeatability and profile scatter must all fit inside that stroke, otherwise the force loop saturates and the chamfer loses uniformity. Third, the trajectories are part-specific: geometry changes require re-importing the model and regenerating the path, although the calibration procedure carries over.
The practical gain is that chamfer quality becomes a property of the programmed process rather than of operator skill. Once the frames are calibrated and the path is validated in simulation, every tooth on every part in the batch is ground under the same contact force, the same tool orientation and the same indexing increment — the condition that makes robotic chamfer grinding of hardened spiral bevel gears viable as a production step rather than a rework station.