Spiral bevel gears transmit motion between intersecting shafts under high torque, high speed, and tight space constraints. Their curved teeth engage gradually, distributing load and reducing impact, but the final grinding operation strongly influences the surface integrity that controls fatigue life. Grinding removes material with thousands of abrasive edges acting almost simultaneously. The same contacts that improve profile accuracy also impose severe mechanical and thermal cycles on a thin surface layer, leaving a residual-stress field that may either resist crack growth or promote it.
The engineering question is not simply whether the ground surface is compressive. It is where the peak compression lies, how far the compressive zone extends, where the stress crosses into tension, and how wheel speed, depth of cut, and feed rate shift that profile. A finite-element analysis of spiral bevel gear grinding, combined with grey relational analysis, provides a way to rank those process variables and to identify a compromise setting within a defined parameter window.
Grinding kinematics and the origin of residual stress
Spiral bevel gear flanks are generated by a virtual producing gear. In a mechanical cradle, the cutter head represents one tooth of that generating gear; in grinding, the wheel performs the same role. The wheel and workpiece roll in mesh without backlash, and the wheel flank sweeps through the tooth space. Contact is therefore a moving, distributed set of abrasive interactions rather than a single pass of an ideal tool.
Each active grain cuts, ploughs, and rubs. Normal and tangential forces produce plastic compression near the surface, while friction and shear generate heat. Heating and cooling are fast, so the surface and subsurface expand and contract against a colder bulk. The final stress state reflects the combined effects of mechanical plastic deformation and thermal strain mismatch. In the model, the workpiece is elastoplastic, the abrasive grains are treated as rigid cones, and a penalty friction law with a coefficient of 0.3 represents the wheel–workpiece interface.
Random multi-abrasive model and material description
A single-grain model cannot capture overlapping scratches, grain interactions, or the statistical distribution of contact. The model therefore uses a virtual lattice: the grinding surface is divided into cells, one conical grain is placed in each cell, and its centre is randomly offset within that cell. The cell size prevents grain overlap, while the offset preserves spatial randomness. The result is a non-uniform multi-abrasive tool without the artificial regularity of a periodic array.
The gear surface is represented by a 2 mm × 1.5 mm × 0.5 mm element. The workpiece is AISI 4340 steel, a high-strength, high-toughness material used for demanding gear applications. The constitutive response uses the Johnson–Cook flow stress with A = 792 MPa, B = 510 MPa, n = 0.26, C = 0.014, and m = 1.03. Failure is described by the Johnson–Cook parameters D1 = 0.05, D2 = 3.44, D3 = −2.12, D4 = 0.002, and D5 = 0.61.
| Parameter | Value |
|---|---|
| Workpiece material | AISI 4340 steel |
| Workpiece density | 7830 kg/m3 |
| Workpiece elastic modulus | 2 × 105 MPa |
| Yield strength | 792 MPa |
| Tensile strength | 980 MPa |
| Workpiece Poisson ratio | 0.29 |
| Workpiece thermal conductivity | 38 W/(m·°C) |
| Abrasive density | 3480 kg/m3 |
| Abrasive elastic modulus | 706 GPa |
| Abrasive Poisson ratio | 0.15 |
| Abrasive thermal conductivity | 79.54 W/(m·°C) |
| Wheel–work friction coefficient | 0.3 |
| Surface element size | 2 × 1.5 × 0.5 mm |
The simulation has two stages. During grinding, the bottom of the workpiece is fixed. After the grinding pass, the tool and contact are removed and cooling boundary conditions are applied to the copied workpiece model. Residual stress is extracted from the S11 component along the grinding direction. For each depth layer, 98 equally spaced nodes are averaged to reduce the influence of edge effects and single-node scatter.
Parameter matrix and stress response
A three-factor, four-level orthogonal matrix is used to cover the process window. Wheel speed vs ranges from 10 to 25 m/s, grinding depth ap from 0.01 to 0.04 mm, and feed speed vw from 3.0 to 4.5 m/min. The 16 runs provide the data for the grey relational ranking and for the response-surface fit.
| Level | vs (m/s) | ap (mm) | vw (m/min) |
|---|---|---|---|
| 1 | 10 | 0.01 | 3.0 |
| 2 | 15 | 0.02 | 3.5 |
| 3 | 20 | 0.03 | 4.0 |
| 4 | 25 | 0.04 | 4.5 |
Across the orthogonal and comparison runs, the peak compressive stress S1 ranges from −235.9 to −396.5 MPa, while the peak tensile stress S2 ranges from 9.1 to 32.0 MPa. In single-factor depth runs at vs = 25 m/s and vw = 4.5 m/min, raising ap from 0.01 to 0.04 mm increases the residual-stress magnitude. The mechanical reason is straightforward: a deeper cut enlarges the contact area, raises the normal force, and increases heat generation. Plastic compression and the thermal gradient both grow. Higher wheel speed and feed act in the same direction within the tested range, but with smaller slopes.
Stress profile along depth
After grinding, the tooth surface is in compression. The maximum compression is not at the outermost node but in the subsurface. The surface experiences the strongest thermal cycle and yields in compression; on cooling, the contraction of the surface layer is constrained by the colder bulk. The combination of a steep thermal gradient and plastic flow places the compressive maximum below the surface. With increasing depth, the stress crosses zero and becomes tensile before decaying toward zero at sufficient depth.
The profiles plotted to about 0.32 mm below the machined surface show this sequence clearly: compressive near the surface, a subsurface compressive peak, a transition to tension, and a gradual return toward equilibrium. This profile matters because the subsurface maximum controls where fatigue cracks may initiate, while the tensile tail can affect distortion and crack growth. A process change that deepens the compressive zone without raising the tensile peak is therefore more desirable than one that simply increases the surface compression.
Grey relational ranking of grinding parameters
The two responses pull in different directions. A setting that increases compressive stress may also increase tensile stress. Grey relational analysis converts the two responses into a single dimensionless score. The maximum compressive stress is normalized as a larger-is-better response in magnitude; the maximum tensile stress is normalized as a smaller-is-better response. A distinguishing coefficient of 0.5 is used, and the grey relational degree G is the average of the two response coefficients.
| Parameter | Level 1 | Level 2 | Level 3 | Level 4 | Range |
|---|---|---|---|---|---|
| vs | 0.5776 | 0.5890 | 0.6117 | 0.6248 | 0.0472 |
| ap | 0.6033 | 0.5764 | 0.5728 | 0.6504 | 0.0776 |
| vw | 0.5842 | 0.5920 | 0.6109 | 0.6159 | 0.0317 |
The range of the level averages is a sensitivity measure. Grinding depth has the largest range, 0.0776, followed by wheel speed at 0.0472 and feed speed at 0.0317. The ranking is therefore ap > vs > vw. The grey-relational optimum is vs = 25 m/s, ap = 0.04 mm, and vw = 4.5 m/min.
The depth level 4 has the highest average score even though it also produces a relatively high tensile response. In the grey score, the compressive coefficient dominates the average. This is an important distinction: the grey relational degree is a weighted composite, not a physical stress, and the selected point is the best compromise under that weighting rather than a point that minimizes tension.
Response-surface model and numerical checks
A second-order response-surface model is fitted to the grey relational degree by least squares, and terms with P > 0.05 are removed. With vs in m/s, ap in mm, and vw in m/min, the predicted grey relational degree Y is
$$
Y = 0.967 – 0.00552v_s + 6.09a_p – 0.2058v_w + 261.1a_p^2 – 0.9072v_s a_p + 0.0084v_s v_w
$$
The model is statistically strong within the tested domain. The analysis of variance gives F = 35.73 and P < 0.0001, with R2 = 95.97% and adjusted R2 = 93.29%. The regression is a surrogate for the grey relational score, not for stress itself, and its validity is limited to the factor ranges used in the matrix.
| Source | DF | SS | MS | F | P |
|---|---|---|---|---|---|
| Regression | 6 | 0.044694 | 0.007449 | 35.73 | < 0.0001 |
| Error | 9 | 0.001876 | 0.000208 | — | — |
| Total | 15 | 0.046570 | — | — | — |
Comparison runs around the optimum expose the trade-off directly. The grey-relational optimum gives S1 = −396.5 MPa and S2 = 32.0 MPa. Reducing wheel speed to 20 m/s lowers the compressive magnitude to −367.0 MPa and the tensile peak to 27.6 MPa. Reducing depth to 0.03 mm gives −372.3 MPa and 29.2 MPa. Reducing feed to 4.0 m/min gives −361.9 MPa and 23.0 MPa. The optimum therefore has the largest compressive magnitude but also the largest tensile peak among these runs.
| Run | vs (m/s) | ap (mm) | vw (m/min) | S1 (MPa) | S2 (MPa) |
|---|---|---|---|---|---|
| 0 | 25 | 0.04 | 4.5 | −396.513 | 32.027 |
| 1 | 20 | 0.04 | 4.5 | −367.031 | 27.616 |
| 2 | 25 | 0.03 | 4.5 | −372.272 | 29.218 |
| 3 | 25 | 0.04 | 4.0 | −361.914 | 23.036 |
The comparison is numerical. It confirms the internal consistency of the ranking, but it also shows that the grey-relational optimum is not a Pareto optimum for tensile stress. If compressive residual stress is the dominant design objective, the selected setting is defensible. If tensile stress, distortion, or crack-growth resistance is the limiting factor, a different point on the response surface should be chosen.
Process implications and limits
Grinding depth is the strongest lever on the residual-stress field. Small changes in ap alter both contact force and heat input, so depth control is the first-order process decision. Wheel speed and feed are secondary but not negligible. Because residual stress increases with all three variables within the tested window, raising productivity will shift the stress profile. The acceptable shift depends on the gear’s dominant fatigue mode, the allowable distortion, and the required subsurface compressive zone.
The model treats grains as rigid cones and omits wheel wear, coolant effects, and a distribution of abrasive shapes. The validation is simulation-based rather than a physical grinding trial. The results are therefore most reliable as a comparative ranking within AISI 4340 and the stated ranges of 10–25 m/s, 0.01–0.04 mm, and 3.0–4.5 m/min. Physical grinding trials and fatigue testing would be needed to translate the ranking into production settings.