Spiral bevel gears transmit motion and torque between intersecting shafts. In the double circular arc variant, the tooth profile is formed by convex and concave circular arcs, producing a line-contact mesh with high load capacity. These gears are used in aerospace, automotive, and precision transmission systems where angular accuracy, noise, and load sharing are closely linked. Transmission error—the deviation between the actual output rotation and the ideal output rotation for a given input—is therefore a key quality indicator. Its measurement is difficult because geometric deviations, mounting errors, and load-induced elastic deformation act together, and because edge contact can distort the coordinate frame used to interpret a measured tooth profile.
Why Transmission Error Is Difficult to Measure
In a double circular arc spiral bevel gear pair, the tooth surface is not a simple involute. The convex and concave arcs form a conformal contact pattern, and the mesh is intended to operate as a line contact rather than a point contact. This gives the gear high load capacity, but it also means that small changes in tooth pitch, tooth alignment, or shaft position can redistribute the contact along the tooth width. Tooth pitch deviation and pinion axis error are especially important: they change the local contact stress and elastic deformation, and they can move the contact zone toward the tooth edges.
Edge contact is a central problem for detection. When the contact moves to an edge, the local contact stress rises, the tooth and housing deform, and the effective position of the gear tooth changes. If the analysis assumes a nominal coordinate frame, the measured profile appears shifted or rotated relative to the theoretical profile. The result is a loss of correspondence between the measured points and the design geometry, which can cause the error search to fail or to return a false deviation. The effect is not a simple linear offset; it depends on load, mounting condition, and the initial contact state.
Tooth alignment modification is used to control this behavior. Common forms include linear end relief, crowning, and helix modification. In each case, material is removed or redistributed along the tooth width so that the load is spread more evenly and the contact does not concentrate at an edge. The relevant parameters include the tilt modification amount, the crowning amount, and the reference length over which the modification is applied. These modifications improve load distribution, but they also change the relationship between the measured profile and the theoretical profile, so a detection method must distinguish intentional modification from unintended transmission error.
Load level changes the picture further. At low torque, the mesh can accommodate part of the geometric error through clearance and local contact adjustment; transmission error tends to decrease when the precision grade is relaxed. At high torque, elastic deformation and load redistribution dominate, and transmission error tends to increase when the precision grade is relaxed. A detection method therefore cannot rely on a single static profile measurement. It must account for the torque-dependent component and for the installation and manufacturing deviations that are already present in the assembled system.
A Coordinate-Frame Error Model
The detection framework begins by establishing a gear coordinate system that allows the actual geometry and position to be compared with the theoretical model. This frame provides a common reference for different gears and for different measurement positions. Within this frame, a controlled-variable analysis is used to separate the influence of torque from the influence of installation and manufacturing factors: one factor is varied while the others are held at fixed conditions, and the resulting change in transmission error is recorded.
The error field is represented by two components, one along the tooth width and one in the radial direction:
$$
e_x = g \exp\left[-k Z \sin\left(\frac{2\pi x}{Z}+2\beta (Z-1)+\delta(x)\right)\right]
$$
$$
e_y = g \exp\left[-k Z \sin\left(\frac{2\pi y}{Z}+2\beta (Z-1)+\delta(y)\right)\right]
$$
Here, \(g\) is the tooth alignment error, \(k\) is the input-torque influence factor, \(Z\) is the number of teeth, \(\beta\) is the helix angle, \(x\) is the position along the tooth width, and \(y\) is the radial position. The functions \(\delta(x)\) and \(\delta(y)\) represent the combined error contribution from tooth-surface contact stress and elastic deformation. The exponential-sinusoidal form gives a periodic variation from tooth to tooth while allowing the torque factor to scale the response. The model is phenomenological rather than a full loaded tooth contact solution: it provides a compact error field for detection and comparison, while the measured profile supplies the actual geometric information.
The model is tied to the meshing geometry through the initial contact position:
$$
A_x = e_x \cos\varphi,\qquad A_y = e_y \sin\varphi
$$
where \(\varphi\) is the rotation angle of the cylindrical gear. From the center-distance variation and the gear rotation, the polar angle and polar radius of the pitch curve can be obtained. The tangent direction of the center-distance curve is then used to project the difference between the actual and theoretical profiles into a scalar transmission error:
$$
\vec{\tau}=d\begin{bmatrix}A_x\cos\varphi-x\\[2pt]A_y\sin\varphi+y\end{bmatrix},\qquad
\hat{\tau}=\frac{\vec{\tau}}{\lVert\vec{\tau}\rVert}
$$
$$
\Delta f=\hat{\tau}\cdot\left(f_2-f_{\text{ideal}}\right)
$$
In these expressions, \(d\) is a constant, \(f_2\) is the theoretical equation of the double circular arc spiral bevel gear, and \(f_{\text{ideal}}\) is the theoretical polar radius at the corresponding rotation angle. The projection converts the two-dimensional profile difference into a single error value along the local normal direction, which is the quantity of interest for transmission error assessment.
Profile Transformation and Normal-Offset Search
The measurement principle follows the double-flank meshing condition. When a standard gear and the test gear mesh on both flanks, the transmission error of the test gear appears as a variation in the center distance. The center distance and the gear rotation angle are therefore used together to locate the corresponding point on the pitch curve and to express the measured profile in the same frame as the theoretical profile.
Measured tooth-profile points are written in homogeneous coordinates:
$$
E_1=\begin{bmatrix}x_i\\y_i\\1\end{bmatrix},\qquad
E_2=\begin{bmatrix}x_j\\y_j\\1\end{bmatrix}
$$
The measured points \(E_1\) are transformed into the theoretical profile frame by \(E_2=N E_1\), using the transformation matrix
$$
N=\begin{bmatrix}
\cos\theta & -\sin\theta & r\\
\sin\theta & \cos\theta & t\\
0 & 0 & 1
\end{bmatrix}
$$
where \(\theta\) is the rotation angle of the coordinate transformation, \(r\) is the gear radius, and \(t\) is the translation time. The rotation and translation remove the rigid-body component of the setup error, so that the remaining normal offset can be interpreted as a profile-related transmission error rather than a misalignment of the measurement frame.
After transformation, the theoretical tooth profile is used to generate normal equidistant lines. Two offset lines are sought: one inward and one outward from the theoretical profile, until both enclose all measured profile points. The minimum enclosing offset represents the profile deviation. This bracket is more robust than matching a single measured point to a single theoretical point, because it accounts for point-to-point noise and for small local deviations along the profile. The search is carried out in two dimensions along the tooth profile, and a piecewise cubic spline is fitted to the resulting offset curve to produce a smooth error function \(H(x,y)\). The smoothed curve is then used to identify the error source and to guide modification or compensation.
The output of this stage is a continuous transmission error curve rather than a set of isolated point deviations. The curve preserves the phase relationship between the measured profile and the theoretical profile, and it allows the effect of load, installation, and tooth alignment modification to be compared on the same basis.
Experimental Configuration
The evaluation used an MMSA spiral bevel gear as the test object. The basic parameters are listed in the table below.
| Parameter | Value |
|---|---|
| Tooth width | 20 mm |
| Profile diameter | 17 mm |
| Profile length | 8.5 mm |
| Outer cone pitch angle | 125° |
| Inner cone pitch angle | 52° |
| Outer cone distance | 80 mm |
| Helix angle | 30° |
| Transmission angle | 10° |
Before measurement, the gear was cleaned and the tooth profile and datum surfaces were wiped. The gear was clamped on a magnetic base, and the probe was calibrated against a standard sphere on the worktable to establish the original position. The gear was then aligned by checking the perpendicularity of the mounting; within the specified range, the gear was considered aligned. A spatial coordinate system was established at the center hole, and the probe was moved slowly and uniformly into contact with the tooth surface. Points were recorded along the planned path, and automatic fitting and path planning were used to complete the sampling.
Profile data were collected in three stages: initial contact, stable operation, and degradation. In the initial contact stage, the contact point is first established, and small changes can be captured with high sampling accuracy. In the stable working stage, the mesh has settled and the basic form remains nearly unchanged; the sampling accuracy is slightly lower than in the initial stage but remains within an acceptable range. In the degradation stage, wear or fault changes the profile shape, so recalibration is needed to maintain data quality. Across these stages, the sampling accuracy was sufficient to support the subsequent transmission error search.
Detection Performance
Under rated load, the transmission error was recorded over a 10–80 ms window and compared with reference values. The proposed method produced a curve that remained close to the reference curve throughout the window. The line-structured-light position-detection method and the RTCP follow-error compensation method showed larger deviations from the reference, particularly where the error curve changed direction. The comparison indicates that the coordinate transformation and normal-offset search were able to preserve the phase and amplitude of the measured error under rated load.
A second evaluation varied the load from 0.0 to 1.2 N·s. As the load increased, the transmission error also increased. Even under this changing load, the proposed method remained closer to the reference than the two comparison methods. This behavior is consistent with the role of the torque influence factor in the error model: the load-dependent elastic deformation changes the error field, but the profile search still recovers the dominant normal deviation.
The cross-arc distance error was used as an additional accuracy indicator. This quantity is the distance error between the two circular arcs in the double circular arc profile. At double-arc angles from 0° to 360°, the relative distance increased with angle, reaching the upper part of the 0–7 mm range. The proposed method followed the reference values more closely than the comparison methods, indicating that the normal-offset search also performs well when the profile geometry changes along the arc.
| Case | Range | Observed trend |
|---|---|---|
| Rated load | 10–80 ms | Proposed curve close to reference; comparison methods deviated |
| Load sweep | 0.0–1.2 N·s | Error increased with load; proposed method remained closest to reference |
| Cross-arc distance | 0–360°; 0–7 mm | Relative distance increased with angle; proposed method matched reference more closely |
Interpretation and Practical Limits
The method combines three elements: a coordinate frame that removes rigid-body setup error, a normal-offset search that brackets the measured profile between two equidistant lines, and a spline-smoothed error curve that can be compared with the theoretical geometry. This combination is useful because transmission error in a double circular arc spiral bevel gear is not a single scalar defect. It is a distributed deviation that changes with load, contact position, and tooth alignment. By expressing the measured points in the theoretical frame before searching, the method reduces the chance that an edge-contact shift is mistaken for a profile error.
The reported evaluation supports the use of the method for high-precision detection under rated load and varying load. However, the evidence is limited to one MMSA gear and to the stated load and time ranges. The error function \(\delta(x),\delta(y)\) is empirical and must be calibrated for a gear family, tooth geometry, and mounting condition. Thermal effects, lubricant film effects, and high-speed dynamic effects are outside the reported tests. The absolute uncertainty of the method also depends on the metrology chain: probe calibration, alignment of the gear axis, sampling density, and the stability of the coordinate transformation.
In practical inspection, the method is best used as part of a closed-loop correction process. The detected error curve can be compared with the intended tooth alignment modification, and the difference can be fed back to the manufacturing or assembly stage. Because the search is tied to the theoretical profile normal, it can distinguish a deliberate crowning or end relief from an unintended transmission error. This makes the approach suitable for acceptance testing, setup compensation, and modification optimization of double circular arc spiral bevel gears, provided that the empirical error model is calibrated and the measurement frame is carefully established.