In my study, I focus on the hard-finishing of internal helical gears because these components are increasingly important in planetary reducers, electric vehicle drivetrains, and compact high-torque transmissions. The gear cutting of internal helical gears with hardened tooth flanks is difficult because the tool accessibility inside the ring is limited, the corrective capability of many existing processes is weak, and the tooth surface topology is not sufficiently flexible. I propose a generating grinding method that uses a spherical worm grinding wheel, and I combine this method with a dedicated grinding wheel dressing strategy. The central idea of my work is that the spherical worm wheel can be dressed accurately by a forming roller, and the dressed wheel can then perform gear cutting of the internal helical gear with improved pitch accuracy, profile accuracy, and topological consistency.
I begin from the geometry of the internal helical gear and the kinematics of a computer numerical control gear grinding machine. The machine has three linear axes and three rotary axes. The linear axes are the radial feed axis, the tangential feed axis, and the axial feed axis. The rotary axes include the tool carrier rotary axis, the worm wheel spindle, the dressing roller spindle, and the workpiece spindle. When I use the spherical worm wheel to grind an internal gear, three main motions are synchronized: the worm wheel rotates about its own axis, the workpiece rotates about its own axis, and the wheel feeds along the tooth width direction. I interpret this process as a meshing motion between an equivalent helical gear and the internal helical gear. This interpretation allows me to derive the wheel surface from the known involute helicoid of the equivalent gear, and it also provides the basis for gear cutting simulation and wheel dressing.
| Symbol | Meaning in my derivation |
|---|---|
| \(S_a\) | Fixed coordinate system of the equivalent helical gear |
| \(S_b\) | Auxiliary coordinate system for center distance |
| \(S_t\) | Moving coordinate system of the equivalent helical gear |
| \(S_c\) | Fixed coordinate system of the worm grinding wheel |
| \(S_w\) | Moving coordinate system of the worm grinding wheel |
| \(\lambda_w\) | Shaft angle parameter between the equivalent gear and the worm wheel |
| \(E_{ws}\) | Shortest distance between the worm wheel axis and the equivalent gear axis |
| \(r_w\) | Radius of the spherical worm grinding wheel |
| \(r_{ps}\) | Radius of the equivalent helical gear |
| \(\phi_s\) | Rotation angle of the equivalent helical gear in the enveloping motion |
| \(\phi_w\) | Rotation angle of the worm grinding wheel in the enveloping motion |
| \(\theta_s\) | Involute development angle |
| \(\theta_1\) | Rotation angle around the gear axis for the helical surface |
| \(p_1\) | Helix parameter of the equivalent helical gear |
| \(r_b\) | Base circle radius of the equivalent helical gear |
| \(\theta_{os}\) | Half angle of the tooth space |
| \(N_t\) | Unit normal vector on the equivalent gear tooth surface |
| \(v_{tw}^{t}\) | Relative velocity between the equivalent gear and the worm wheel in \(S_t\) |
| \(f_{ws}\) | Meshing condition function |
For my numerical example, I use an internal helical gear with the parameters listed in the table below. These parameters define the target tooth surface and are used repeatedly in the enveloping calculation, the virtual gear cutting simulation, and the experimental validation. I keep the same parameter set throughout the study so that the effect of wheel dressing can be isolated from other factors.
| Parameter | Value |
|---|---|
| Number of teeth \(Z\) | 79 |
| Normal module \(m_n\) | 2 mm |
| Transverse module \(m_t\) | 2.071 mm |
| Normal pressure angle \(\alpha_n\) | 20 degrees |
| Helix angle \(\beta\) | 15 degrees |
| Face width \(b\) | 45 mm |
| Hand of helix | Left |
To describe the enveloping relationship, I build the coordinate systems shown logically in my kinematic model. The transformation from the moving equivalent gear system to the fixed worm wheel system is obtained by multiplying the homogeneous transformation matrices in sequence. I write the total transformation as
$$M_{wt}=M_{wc}M_{cb}M_{ba}M_{at}.$$
Each individual matrix represents one spatial relation. The transformation from the equivalent gear moving frame to the fixed gear frame is a rotation about the gear axis, and I write it as
$$M_{at}=\begin{bmatrix}
\cos\phi_s & -\sin\phi_s & 0 & 0\\
\sin\phi_s & \cos\phi_s & 0 & 0\\
0 & 0 & 1 & 0\\
0 & 0 & 0 & 1
\end{bmatrix}.$$
The offset between the two gear axes is represented by the center distance. In my coordinate construction, I write this offset matrix as
$$M_{ba}=\begin{bmatrix}
1 & 0 & 0 & -E_{ws}\\
0 & 1 & 0 & 0\\
0 & 0 & 1 & 0\\
0 & 0 & 0 & 1
\end{bmatrix}.$$
The shaft angle relation is included through a rotation in the auxiliary frame. I use
$$M_{cb}=\begin{bmatrix}
1 & 0 & 0 & 0\\
0 & \cos(\lambda_w+\pi/2) & \sin(\lambda_w+\pi/2) & 0\\
0 & \sin(\lambda_w+\pi/2) & \cos(\lambda_w+\pi/2) & 0\\
0 & 0 & 0 & 1
\end{bmatrix}.$$
The rotation of the worm wheel in its own moving frame is expressed as
$$M_{wc}=\begin{bmatrix}
\cos\phi_w & \sin\phi_w & 0 & 0\\
-\sin\phi_w & \cos\phi_w & 0 & 0\\
0 & 0 & 1 & 0\\
0 & 0 & 0 & 1
\end{bmatrix}.$$
Because the spherical worm wheel surface is complex, I do not attempt to define it directly from a simple tool profile. Instead, I derive it by reversing the enveloping process from the equivalent helical gear. The equivalent gear is a standard involute helical gear. Its tooth surface can be written in parametric form as
$$r_t(\theta_s,\theta_1)=\begin{bmatrix}x_t\\y_t\\z_t\\1\end{bmatrix},$$
where the coordinate components are
$$x_t=r_b\cos(\theta_s+\theta_1+\theta_{os})+r_b\theta_s\sin(\theta_s+\theta_1+\theta_{os}),$$
$$y_t=\pm r_b\sin(\theta_s+\theta_1+\theta_{os})\mp r_b\cos(\theta_s+\theta_1+\theta_{os}),$$
$$z_t=p_1\theta_1.$$
In these expressions, the involute geometry is controlled by the base radius, the development angle, and the helix parameter. The parameter \(\theta_1\) produces the axial shift of the helical surface, while \(\theta_s\) controls the position along the involute profile. The sign choice distinguishes the two flanks of the tooth. I then generate the worm wheel surface family by applying the total transformation to the equivalent gear surface:
$$r_w(\theta_s,\theta_1,\phi_s)=M_{wt}r_t(\theta_s,\theta_1).$$
The enveloping condition is that the relative velocity at a contact point must be tangent to both surfaces. Mathematically, I enforce the meshing equation
$$f_{ws}(\theta_s,\theta_1,\phi_s)=N_t\cdot v_{tw}^{t}=0,$$
where \(N_t\) is the unit normal vector of the equivalent gear tooth surface and \(v_{tw}^{t}\) is the relative velocity expressed in the equivalent gear moving frame. The unit normal is obtained from the cross product of the partial derivatives of the tooth surface:
$$N_t=\frac{\frac{\partial r_t}{\partial \theta_s}\times \frac{\partial r_t}{\partial \theta_1}}{\left|\frac{\partial r_t}{\partial \theta_s}\times \frac{\partial r_t}{\partial \theta_1}\right|}=\begin{bmatrix}
\sin(\theta_{os}+\theta_s)\\
\pm \cos(\theta_{os}+\theta_s)\\
0
\end{bmatrix}.$$
The velocity of a point on the equivalent gear surface is
$$v_t^{t}=\omega_t^{t}\times r_t,$$
and the velocity of the corresponding point on the worm wheel surface, expressed in the equivalent gear frame, is
$$v_w^{t}=\omega_w^{t}\times r_w.$$
Therefore, the relative velocity becomes
$$v_{tw}^{t}=v_t^{t}-v_w^{t}=\left(\omega_t^{t}-\omega_w^{t}\right)\times r_t+E\times \omega_w^{t}.$$
The offset vector between the surfaces in the equivalent gear frame is written as
$$E=r_w-r_t=\begin{bmatrix}
\cos\phi_s & \sin\phi_s & 0\\
-\sin\phi_s & \cos\phi_s & 0\\
0 & 0 & 0
\end{bmatrix}\begin{bmatrix}E_{ws}\\0\\0\end{bmatrix}.$$
The angular velocity of the worm wheel, expressed in the equivalent gear frame, is
$$\omega_w^{t}=\begin{bmatrix}
\cos\phi_s & \sin\phi_s & 0\\
-\sin\phi_s & \cos\phi_s & 0\\
0 & 0 & 1
\end{bmatrix}\begin{bmatrix}
1 & 0 & 0\\
0 & \cos(\lambda_w+90^\circ) & \sin(\lambda_w+90^\circ)\\
0 & -\sin(\lambda_w+90^\circ) & \cos(\lambda_w+90^\circ)
\end{bmatrix}\begin{bmatrix}0\\0\\\omega_t\end{bmatrix}.$$
The angular velocity of the equivalent gear is simply
$$\omega_t^{t}=\begin{bmatrix}0\\0\\\omega_t\end{bmatrix}.$$
After substituting the surface coordinates, the offset vector, and the angular velocities into the relative velocity expression, I obtain the three components of \(v_{tw}^{t}\). I write them as
$$v_{tw}^{t}=\begin{bmatrix}
\omega_w z_t\cos\phi_s\cos\lambda_w-y_t\left(\omega_w(-\sin\lambda_w)-\omega_t\right)+\omega_w E_{ws}\sin\phi_s\sin\lambda_w\\
x_t\left(\omega_w(-\sin\lambda_w)-\omega_t\right)-\omega_w z_t\sin\phi_s\cos\lambda_w+\omega_w E_{ws}\cos\phi_s\sin\lambda_w\\
\omega_w\cos\lambda_w\left(y_t\sin\phi_s-x_t\cos\phi_s+E_{ws}\right)
\end{bmatrix}.$$
Combining this relative velocity with the unit normal gives the explicit meshing equation:
$$f_{ws}(\theta_s,\theta_1,\phi_s)=\sin(\theta_{os}+\theta_s)\left[\omega_w z_t\cos\phi_s\cos\lambda_w-y_t\left(\omega_w(-\sin\lambda_w)-\omega_t\right)+\omega_w E_{ws}\sin\phi_s\sin\lambda_w\right]-\cos(\theta_{os}+\theta_s)\left[x_t\left(\omega_w(-\sin\lambda_w)-\omega_t\right)-\omega_w z_t\sin\phi_s\cos\lambda_w+\omega_w E_{ws}\cos\phi_s\sin\lambda_w\right].$$
For a given pair of parameters \(\theta_1\) and \(\phi_s\), the meshing equation is solved for \(\theta_s\). This solution defines the contact line on the equivalent gear surface that generates the worm wheel surface. I then substitute the solved relation \(\theta_s=\theta_s(\theta_1,\phi_s)\) into the surface family and obtain the worm wheel tooth surface as
$$R_w(\theta_1,\phi_s)=r_w(\theta_1,\theta_s(\theta_1,\phi_s),\phi_s).$$
This surface is the theoretical spherical worm wheel surface that I aim to produce by dressing. In my view, the accuracy of gear cutting with this wheel depends strongly on how closely the dressed wheel surface matches this theoretical surface. Therefore, I next analyze the dressing method.
| Worm wheel parameter | Value |
|---|---|
| Number of worm starts \(N_t\) | 79 |
| Normal module | 2 mm |
| Pressure angle \(\alpha_n\) | 20 degrees |
| Profile wheel normal module | 2 mm |
| Profile wheel helix angle \(\beta_1\) | 10 degrees |
| Profile wheel number of teeth \(N_w\) | 25 |
| Center distance \(E_{ws}\) | 50 mm |
Grinding wheel dressing is a decisive step in my process because the dressed wheel transfers its geometry to the internal gear during gear cutting. Two common dressing strategies are point dressing and forming dressing. For a spherical worm wheel with a complex, continuously curved surface, point dressing is not practical because the contact area is too small and the wheel profile cannot be corrected efficiently across the full tooth height. I therefore select forming dressing. In forming dressing, the diamond roller reproduces the required profile on the wheel surface, and the relative motion between the roller and the wheel is controlled by the machine axes.
I use the virtual center distance principle to design the dressing motion. The principle states that an arbitrary planar motion of a rigid body can be decomposed into a rotation about a selected base point and a translation of that base point. In the actual machine, the diamond roller is usually fixed and cannot be swung along a circular arc by a dedicated axis. Therefore, I convert the required roller motion into the motion of the spherical worm wheel. The eccentric swing of the wheel is decomposed into two parts: a translation in the tangential-axial plane and a rotation about the tool carrier axis. The translation is realized by the tangential feed axis and the axial feed axis, while the rotation is realized by the tool carrier rotary axis.
I express the virtual center distance motion in a compact form. If the swing angle of the wheel about the virtual center is \(\phi_A\), and the virtual radius is \(R_v\), then the tangential displacement and axial displacement can be written as
$$y_A(t)=R_v\sin\phi_A(t),$$
$$z_A(t)=R_v\left(1-\cos\phi_A(t)\right).$$
Conversely, for a commanded tangential displacement, the required swing angle is
$$\phi_A(t)=\arcsin\left(\frac{y_A(t)}{R_v}\right),$$
and the axial displacement follows as
$$z_A(t)=R_v-\sqrt{R_v^2-y_A(t)^2}.$$
These relations define the synchronous motion of the tangential axis, axial axis, and tool carrier axis during dressing. The rotary axis of the diamond roller also rotates to maintain the proper cutting speed. In my dressing scheme, the roller and the worm wheel rotate in a coordinated manner so that the roller envelope sweeps the desired spherical worm surface. The gear cutting performance of the wheel is then determined by the accuracy of this envelope.
| Machine axis | Function in my process | Role in dressing |
|---|---|---|
| Radial axis \(x\) | Controls infeed for gear cutting | Sets dressing depth and center distance |
| Tangential axis \(y\) | Feeds the wheel along tooth width | Provides part of the virtual center translation |
| Axial axis \(z\) | Feeds the wheel along the gear axis | Provides the remaining virtual center translation |
| Tool carrier axis \(A\) | Orients the wheel relative to the workpiece | Provides the eccentric swing |
| Worm wheel spindle \(B_1\) | Rotates the spherical worm wheel | Generates the dressing envelope |
| Dressing roller spindle \(B_2\) | Rotates the diamond roller | Removes material from the wheel |
| Workpiece spindle \(C\) | Rotates the internal gear | Not active during dressing, but synchronized in gear cutting |
I verify the dressing theory by numerical simulation. I calculate the theoretical spherical worm wheel surface from the enveloping equations, and I separately simulate the surface generated by the virtual center distance dressing motion. I then compare the two surfaces along selected helical lines on the left and right flanks. The deviation at a sample point is defined as the projection of the coordinate difference onto the local surface normal:
$$\delta_i=\left(\mathbf{R}_{sim,i}-\mathbf{R}_{th,i}\right)\cdot \mathbf{n}_i.$$
Here \(\mathbf{R}_{sim,i}\) is the simulated dressed surface point, \(\mathbf{R}_{th,i}\) is the theoretical surface point, and \(\mathbf{n}_i\) is the local unit normal. In my numerical evaluation, I select eight helical lines on each flank and calculate the normal deviation along each line. The maximum deviation remains below \(6\times 10^{-6}\) mm. This value is far below the practical tolerance of the gear grinding process and confirms that the virtual center distance motion can envelope the correct worm wheel profile. The deviation grows slightly from the central region of the wheel toward the two ends, and the left and right flank deviations have opposite signs. This behavior is consistent with the symmetry of the spherical wheel and with the small numerical approximation used in the enveloping solution.
| Sampled helical line | Left flank deviation \((\times 10^{-6}\,\text{mm})\) | Right flank deviation \((\times 10^{-6}\,\text{mm})\) |
|---|---|---|
| 1 | -5.8 | 5.6 |
| 2 | -4.9 | 4.8 |
| 3 | -4.1 | 3.9 |
| 4 | -3.2 | 3.0 |
| 5 | -2.1 | 2.0 |
| 6 | -1.0 | 0.9 |
| 7 | 0.8 | -0.7 |
| 8 | 1.5 | -1.4 |
The numerical result supports my claim that the dressing motion is kinematically correct. However, a correct kinematic model alone does not guarantee a correct gear cutting result. The dressed wheel must also produce the intended internal gear surface under the actual generating motion. Therefore, I perform a virtual machining simulation. In this simulation, I model the internal helical gear blank, the spherical worm wheel, and the machine axis motions. I then simulate the material removal process in different grinding stages. The purpose is to compare the simulated gear surface with the theoretical gear surface and to evaluate overcut and residual material.
In the virtual gear cutting simulation, I define the tool path from the same spatial meshing relation used in the theoretical derivation. The wheel rotates about its own axis, the workpiece rotates about its own axis, and the feed motion advances the contact line along the tooth width. At each simulation step, the wheel surface intersects the gear blank and removes material. I use a solid modeling environment to track the removed volume and to extract the final tooth surface. The comparison between the simulated surface and the theoretical surface is performed on the same flank grids.
| Simulation stage | Cutting depth | Observed surface condition |
|---|---|---|
| Initial contact | 0 mm | The wheel touches the blank without material removal |
| Partial profile cutting | 0.2 mm | The tooth flank begins to form, with visible uncut regions |
| Deeper profile cutting | 0.5 mm | The flank shape approaches the theoretical involute |
| Final generating pass | Completed | The simulated gear matches the theoretical gear in most regions |
When I compare the results before and after wheel dressing, the difference is clear. Before dressing, the simulated gear shows larger overcut regions and residual material, especially near the tooth root and tooth tip. These errors are caused by the geometric deviation of the undressed wheel from the required spherical worm surface. After dressing, the overcut and residual errors become localized and small. Most of the tooth surface coincides with the theoretical surface, and only a small amount of residual error remains in the transition zones. This confirms that the dressing method improves the geometric fidelity of the wheel and therefore improves the gear cutting accuracy.
| Comparison item | Before wheel dressing | After wheel dressing |
|---|---|---|
| Overall surface matching | Visible mismatch on both flanks | Close match over most of the tooth surface |
| Overcut distribution | Large and distributed | Small and localized |
| Residual material | Large near root and tip | Small in transition zones |
| Expected gear cutting accuracy | Limited by wheel profile error | Improved by corrected wheel profile |
To validate the virtual result in practice, I conduct grinding experiments on a computer numerical control gear grinding machine. I first dress the spherical worm wheel with a diamond roller using the virtual center distance motion. Then I use the dressed wheel to perform gear cutting of the internal helical gear. For comparison, I also grind a gear without wheel dressing under otherwise identical conditions. I measure both gears on a gear measuring platform and evaluate pitch deviation, cumulative pitch deviation, and profile deviation. These inspections provide direct evidence of the effect of wheel dressing on gear cutting.
The measurement results show that the dressed wheel produces a more accurate internal helical gear. The single pitch deviation of the left flank decreases from 4.4 \(\mu\)m to 3.6 \(\mu\)m, and the single pitch deviation of the right flank decreases from 7.9 \(\mu\)m to 5.9 \(\mu\)m. The cumulative pitch deviation of the left flank decreases from 27.4 \(\mu\)m to 17.2 \(\mu\)m, while the cumulative pitch deviation of the right flank decreases from 28.8 \(\mu\)m to 27.1 \(\mu\)m. The total profile deviation decreases from 8.4 \(\mu\)m to 4.9 \(\mu\)m. In terms of accuracy grade, the right cumulative pitch and right single pitch improve from grade 6 to grade 5, the left cumulative pitch improves from grade 6 to grade 5, the left single pitch improves from grade 5 to grade 4, and the profile total deviation improves from grade 7 to grade 5. These improvements demonstrate that wheel dressing is not a minor correction; it is a necessary condition for high-quality gear cutting with the spherical worm wheel.
| Internal gear error | Without wheel dressing | With wheel dressing | Reduction | Accuracy grade change |
|---|---|---|---|---|
| Right cumulative pitch deviation \(F_p\) | 28.8 \(\mu\)m | 27.1 \(\mu\)m | 5.9% | Grade 6 to grade 5 |
| Left cumulative pitch deviation \(F_p\) | 27.4 \(\mu\)m | 17.2 \(\mu\)m | 37.2% | Grade 6 to grade 5 |
| Right single pitch deviation \(f_p\) | 7.9 \(\mu\)m | 5.9 \(\mu\)m | 25.3% | Grade 6 to grade 5 |
| Left single pitch deviation \(f_p\) | 4.4 \(\mu\)m | 3.6 \(\mu\)m | 18.2% | Grade 5 to grade 4 |
| Total profile deviation \(F_\alpha\) | 8.4 \(\mu\)m | 4.9 \(\mu\)m | 39.3% as evaluated | Grade 7 to grade 5 |
For a clearer interpretation, I define the percentage reduction of an error as
$$\eta_e=\frac{e_{\text{without}}-e_{\text{with}}}{e_{\text{without}}}\times 100\%.$$
Using this definition, the left single pitch deviation decreases by 18.2%, the right single pitch deviation decreases by 25.3%, and the left cumulative pitch deviation decreases by 37.2%. The right cumulative pitch deviation decreases by 5.9%, which is a smaller relative reduction but still significant because the absolute value remains controlled. The profile total deviation is reduced by approximately 39.3% according to the inspection evaluation. The difference in improvement between the left and right flanks is caused by the asymmetric contact condition inside the internal gear, the local wear of the dressing roller, and the small machine axis deviations during generating motion. Even so, every measured error is reduced, and every accuracy grade either improves or remains favorable.
I also examine the relationship between the dressed wheel surface and the final gear surface. During gear cutting, the wheel surface is not copied directly onto the gear; instead, the gear surface is the envelope of the wheel surface along the generating motion. Therefore, a small dressing error can be magnified or attenuated depending on the local contact geometry. The virtual machining comparison shows that the largest residual errors after dressing occur in areas where the contact line changes rapidly. These areas are near the tooth root fillet and near the top edge of the flank. In practical gear cutting, I can reduce these residual errors further by optimizing the dressing roller profile, increasing the number of dressing passes, and compensating the axis synchronization errors. My current results already show that the spherical worm wheel with dressing can produce a grade 5 profile and improved pitch accuracy, which is sufficient for many internal gear applications.
The influence of the virtual center distance on the dressing result can be summarized by the following relations. If the dressing roller radius is \(R_r\), the wheel radius is \(R_w\), and the desired contact angle is \(\alpha_d\), then the equivalent center distance for the dressing motion must satisfy
$$E_d=R_w+R_r-2\sqrt{R_w R_r}\cos\alpha_d.$$
For the spherical worm wheel, the dressing contact is not a simple line contact but a complex spatial contact. I therefore use the virtual center distance to map the roller motion into the machine axes. The mapping is
$$\begin{bmatrix}
\Delta y_d\\
\Delta z_d\\
\Delta A_d
\end{bmatrix}
=
\begin{bmatrix}
R_v\sin\phi_A\\
R_v(1-\cos\phi_A)\\
\phi_A
\end{bmatrix}.$$
By differentiating this mapping with respect to time, I obtain the velocity relation used in the interpolation algorithm:
$$\begin{bmatrix}
\dot{y}_d\\
\dot{z}_d\\
\dot{A}_d
\end{bmatrix}
=
\begin{bmatrix}
R_v\cos\phi_A\\
R_v\sin\phi_A\\
1
\end{bmatrix}\dot{\phi}_A.$$
This velocity relation is important because any lag between the tangential axis, axial axis, and tool carrier axis will produce a dressing error on the wheel. In my experiments, I ensure that the servo drives of these axes are synchronized through the electronic gearbox. The diamond roller spindle is speed-controlled separately, but its rotational phase is monitored to avoid mismatch with the wheel spindle. The wheel spindle and the roller spindle are coupled through the desired dressing speed ratio:
$$n_w=\frac{N_r}{N_w}n_r,$$
where \(n_w\) is the wheel spindle speed, \(n_r\) is the roller spindle speed, \(N_r\) is the number of effective roller starts, and \(N_w\) is the number of wheel starts. This ratio maintains the correct envelope speed and prevents local burning or excessive wear.
| Dressing variable | Definition | Effect on gear cutting |
|---|---|---|
| Virtual radius \(R_v\) | Radius of the decomposed eccentric swing | Controls the curvature of the dressed surface |
| Swing angle \(\phi_A\) | Rotation angle about the virtual center | Controls the tangential and axial motion amplitudes |
| Roller speed \(n_r\) | Rotational speed of the dressing roller | Controls dressing efficiency and wheel topography |
| Wheel speed \(n_w\) | Rotational speed of the worm wheel | Controls the envelope density on the wheel surface |
| Dressing depth | Radial infeed of the roller | Controls wheel profile size and sharpness |
| Axis synchronization error | Deviation among \(y\), \(z\), and \(A\) | Directly creates profile error in gear cutting |
After gear cutting, I also inspect the tooth surface topography. The dressed wheel produces a more uniform contact pattern along the tooth width and tooth height. The undressed wheel leaves visible generating marks and local deviations, especially where the wheel profile deviates from the theoretical surface. The dressed wheel reduces these marks and makes the tooth surface more consistent with the theoretical involute helicoid. This is important for internal gears because the contact conditions inside the ring are often less stable than those of external gears, and any local profile error can lead to noise, vibration, and uneven load distribution.
In my opinion, the main contribution of this work is the combination of a spherical worm wheel model, a virtual center distance dressing method, and a full validation chain from numerical simulation to virtual gear cutting to experimental measurement. The mathematical derivation gives the theoretical wheel surface. The dressing simulation proves that the wheel can be generated by the proposed axis motion. The virtual machining simulation proves that the dressed wheel produces the intended gear surface. The measurement report proves that the actual gear cutting accuracy improves. This chain is necessary because internal gear generating grinding is a coupled process: the wheel geometry, the dressing motion, and the generating motion all influence the final tooth surface.
The limitations of my current work should also be noted. The dressing roller profile is assumed to be ideal, and the machine axis errors are not fully modeled. The thermal deformation of the wheel and workpiece is not included in the simulation. The wear of the diamond roller is treated as a secondary effect, although in long production runs it can become important. The virtual center distance method requires accurate synchronization of three axes, and any servo mismatch will create a periodic dressing error. In future work, I plan to include roller wear compensation, thermal error compensation, and closed-loop measurement feedback into the gear cutting process. I also plan to extend the method to different helix angles and different internal gear sizes.
| Research stage | Method I use | Output | Validation criterion |
|---|---|---|---|
| Theoretical wheel design | Involute gear enveloping and homogeneous transformations | Spherical worm wheel surface equation | Meshing condition is satisfied |
| Dressing kinematics | Virtual center distance decomposition | Axis motion relations for \(y\), \(z\), and \(A\) | Dressed surface matches theoretical surface |
| Numerical simulation | MATLAB surface comparison | Deviation along selected helical lines | Maximum deviation below \(6\times 10^{-6}\) mm |
| Virtual gear cutting | Solid-based material removal simulation | Simulated internal gear model | Overcut and residual material are localized |
| Experimental grinding | CNC gear grinding with and without dressing | Ground internal helical gears | Pitch and profile errors are reduced |
| Accuracy inspection | Gear measuring platform | Pitch, cumulative pitch, and profile reports | Accuracy grade improves |
I summarize the numerical and experimental improvements in another form. Let \(E_0\) be an error without dressing and \(E_1\) be the same error with dressing. The improvement ratio is
$$I=\frac{E_0-E_1}{E_0}.$$
For the left single pitch deviation,
$$I_{f_p,L}=\frac{4.4-3.6}{4.4}=0.182=18.2\%.$$
For the right single pitch deviation,
$$I_{f_p,R}=\frac{7.9-5.9}{7.9}=0.253=25.3\%.$$
For the left cumulative pitch deviation,
$$I_{F_p,L}=\frac{27.4-17.2}{27.4}=0.372=37.2\%.$$
For the right cumulative pitch deviation,
$$I_{F_p,R}=\frac{28.8-27.1}{28.8}=0.059=5.9\%.$$
For the profile total deviation, the inspection evaluation gives an improvement of about 39.3%. These ratios confirm that the dressed wheel improves both the discrete pitch accuracy and the continuous profile accuracy. The stronger improvement on the left flank suggests that the dressing correction is more effective for the flank that experiences a more favorable contact condition. The right flank still improves, but the improvement is limited by the remaining machine and setup errors.
I also express the overall gear cutting error as a combination of several contributions:
$$E_{\text{total}}^2=E_{\text{wheel}}^2+E_{\text{dress}}^2+E_{\text{motion}}^2+E_{\text{thermal}}^2+E_{\text{fixture}}^2.$$
In this model, \(E_{\text{wheel}}\) is the theoretical wheel surface error, \(E_{\text{dress}}\) is the dressing error, \(E_{\text{motion}}\) is the generating motion error, \(E_{\text{thermal}}\) is the thermal error, and \(E_{\text{fixture}}\) is the workpiece clamping error. Wheel dressing mainly reduces \(E_{\text{wheel}}\) and \(E_{\text{dress}}\). The virtual center distance method reduces \(E_{\text{dress}}\) by controlling the roller envelope. The generating motion error remains important, but once the wheel is dressed correctly, the gear cutting result is no longer dominated by the wheel profile error. This explains why the pitch and profile deviations decrease after dressing.
The gear cutting process with the spherical worm wheel can be described as a sequence of transformations from the tool surface to the workpiece surface. If \(S_w\) is the wheel surface and \(T(t)\) is the generating transformation at time \(t\), then the generated gear surface is the envelope of the family
$$S_g=\text{env}_{t}\left[T(t)S_w\right].$$
The envelope condition is that the relative velocity between the wheel and the workpiece is tangent to the contact point. This condition is the same as the meshing equation used in the wheel design, but it is applied in the opposite direction. In wheel design, the equivalent gear generates the wheel. In gear cutting, the wheel generates the internal gear. The consistency of the two envelopes is the reason the method works. If the wheel is dressed according to the correct envelope, the gear cutting envelope reproduces the desired internal gear surface.
For the internal helical gear, the tooth surface can be represented in the workpiece coordinate system as
$$r_g(u,v)=\begin{bmatrix}
r_b\cos(u+v+\theta_0)+r_b u\sin(u+v+\theta_0)\\
\pm r_b\sin(u+v+\theta_0)\mp r_b\cos(u+v+\theta_0)\\
p_g v
\end{bmatrix},$$
where \(u\) is the involute parameter, \(v\) is the axial parameter, \(r_b\) is the base radius, \(p_g\) is the helix parameter, and \(\theta_0\) is the angular position. The generating motion must satisfy
$$N_g\cdot v_{gw}=0.$$
This condition ensures that the wheel and the internal gear contact along the generating line. In my experiments, the deviation between the measured surface and this theoretical surface is reduced after dressing, which confirms that the actual generating contact is closer to the ideal condition.
I also compare the measured pitch deviations with the theoretical limits. The single pitch deviation describes the difference between adjacent tooth spaces, and the cumulative pitch deviation describes the accumulation of these differences around the gear. The definitions can be written as
$$f_p=\max_i\left|p_i-p_{\text{nom}}\right|,$$
$$F_p=\max_i\left(P_i\right)-\min_i\left(P_i\right),$$
where \(p_i\) is the actual pitch, \(p_{\text{nom}}\) is the nominal pitch, and \(P_i\) is the cumulative pitch at tooth \(i\). The profile total deviation is defined as
$$F_\alpha=\max_i\left|d_i\right|,$$
where \(d_i\) is the departure of the measured profile from the design profile. These definitions are standard in gear inspection, and they allow me to compare the effect of wheel dressing quantitatively. The reduction in \(f_p\) shows that the indexing motion and tooth spacing are improved. The reduction in \(F_p\) shows that the accumulated error around the gear is reduced. The reduction in \(F_\alpha\) shows that the involute profile is more accurate.
| Error definition | Formula | Interpretation in my gear cutting test |
|---|---|---|
| Single pitch deviation | \(f_p=\max_i\left|p_i-p_{\text{nom}}\right|\) | Adjacent tooth spacing error |
| Cumulative pitch deviation | \(F_p=\max_i P_i-\min_i P_i\) | Accumulated indexing error |
| Profile total deviation | \(F_\alpha=\max_i\left|d_i\right|\) | Involute profile form error |
| Normal surface deviation | \(\delta_i=\left(\mathbf{R}_{sim,i}-\mathbf{R}_{th,i}\right)\cdot \mathbf{n}_i\) | Wheel dressing or virtual machining surface error |
The gear cutting experiments reveal another important effect: the dressed wheel not only reduces the average error but also reduces the variation of the error along the tooth width. Before dressing, the pitch and profile errors change significantly from one tooth to another. After dressing, the error curves become smoother. This is beneficial for internal gears because the ring structure is sensitive to local load concentrations. A smoother tooth surface distributes the load more evenly and reduces the risk of edge contact. The improved surface also reduces the excitation that can cause noise and vibration in planetary gear sets.
From a manufacturing perspective, the dressing process adds a small amount of cycle time, but the benefit in gear cutting accuracy is substantial. The dressing operation is performed intermittently, and one dressed wheel can grind multiple internal gears before redressing is required. The virtual center distance method does not require a special roller swing axis, so it can be implemented on a standard multi-axis gear grinding machine with coordinated linear and rotary axes. This makes the method practical for production. I believe that the combination of spherical worm wheel generating grinding and virtual center distance dressing is a strong candidate for high-precision gear cutting of internal helical gears.
| Manufacturing factor | Without dressing | With virtual center distance dressing |
|---|---|---|
| Wheel profile accuracy | Poor due to undressed spherical surface | Improved by controlled roller envelope |
| Gear cutting accuracy | Limited by wheel profile error | Improved pitch and profile grades |
| Process flexibility | Low for complex internal gears | High because wheel form can be corrected |
| Machine requirement | Standard multi-axis machine | Standard multi-axis machine with synchronized axes |
| Production stability | Error varies with wheel wear | Error can be controlled by periodic dressing |
| Application potential | Limited for hardened internal gears | Strong for electric vehicle planetary reducers |
In conclusion, I have studied a generating grinding method for internal helical gears using a spherical worm wheel and a forming dressing strategy. I derived the worm wheel surface by solving the enveloping conditions between the wheel and an equivalent helical gear. I formulated the homogeneous transformation matrices, the unit normal vector, the relative velocity, and the meshing equation. I then designed a virtual center distance dressing method that decomposes the required roller motion into machine axis motions. The numerical simulation shows that the dressed surface deviates from the theoretical surface by less than \(6\times 10^{-6}\) mm. The virtual gear cutting simulation shows that the dressed wheel produces only small localized overcut and residual errors compared with the undressed wheel. The experimental gear cutting results show that the left and right single pitch deviations decrease by 18.2% and 25.3%, the left and right cumulative pitch deviations decrease by 37.2% and 5.9%, and the profile total deviation decreases by about 39.3%. The accuracy grades improve accordingly. These results confirm that wheel dressing is essential for accurate gear cutting of internal helical gears with a spherical worm wheel, and that the virtual center distance method is a feasible and effective dressing solution.
I plan to continue this work by improving the dressing roller design, adding closed-loop compensation for axis synchronization, and extending the model to include thermal and wear effects. I also intend to apply the method to a wider range of internal gear geometries and to optimize the gear cutting parameters for shorter cycle time and longer wheel life. The results presented here provide a foundation for high-efficiency, high-precision gear cutting of internal helical gears, and they demonstrate that the spherical worm wheel generating method can be a practical alternative to conventional form grinding and honing for hardened internal gear surfaces.