In the field of mechanical power transmission, spur gears are among the most fundamental and widely used components due to their simplicity and efficiency. However, the inherent dynamics of gear meshing, such as transmission error, vibration, and noise, pose significant challenges to high-performance applications. Gear modification, which involves intentional deviations from the ideal tooth profile, has been established as a critical technique to mitigate these issues, enhance load distribution, and improve overall reliability. While profile modification or lead crowning have been studied extensively in isolation, the synergistic effects of combined profile and lead modification—termed compound modification—present a more comprehensive solution, yet one that is complex to optimize. Traditional optimization approaches often require extensive computational resources and iterative manual adjustments, and many dynamic analyses overlook the influence of supporting structures like bearings, shafts, and gyroscopic effects. This work introduces a novel, efficient design methodology for the compound modification of spur gears, integrating finite element analysis, a finite element node-based dynamic model, and the Kriging surrogate modeling technique to achieve optimal gear performance with consideration of real-world system dynamics.
The core principle of compound modification for spur gears lies in simultaneously applying profile modification along the tooth flank and lead modification across the tooth width. Profile modification primarily addresses excitations caused by machining errors, elastic deformations under load, and the resulting meshing impacts at the entry and exit points of contact. A common representation for the profile modification curve is a power function. Let the modification amount be denoted by $C_{\alpha}$, the length of modification along the line of action be $l$, and the relative coordinate from the start of modification be $x$. The modification depth $e$ at any point is given by:
$$ e = C_{\alpha} \left( \frac{x}{l} \right)^b $$
where $b$ is the power exponent defining the shape of the modification curve. The parameters for profile modification thus include the modification amount $C_{\alpha}$, the modification length (often expressed as a multiple of the module, $L_{Ca}$), and the exponent $b$.
Lead modification, typically in the form of crowning, is applied to compensate for misalignments, shaft deflections, and manufacturing errors that can cause edge loading and stress concentration. The lead crowning is usually symmetric, with the maximum crown amount $C_{\beta}$ at the center of the tooth width tapering to zero at the ends. The combined effect of these two modifications on the tooth surface geometry is complex, and their parameters must be carefully coordinated to avoid detrimental interactions and to maximize beneficial outcomes such as reduced dynamic transmission error (DTE) and contact stress.
To accurately capture the dynamic behavior of a spur gear pair within a mechanical system, a comprehensive dynamic model is essential. The finite element node method is employed here, which models the gear-shaft-bearing system as a flexible rotor system. This method offers significant advantages: it accounts for the flexibility of the shafts, the stiffness and damping of the bearings, and the gyroscopic effects of rotating components—factors often simplified or ignored in lumped-parameter models. The system is discretized using nodes along the shafts, each possessing six degrees of freedom (three translational and three rotational). For a single-stage spur gear pair, the overall system dynamics can be derived by assembling the equations for individual components.
The equation of motion for the gear rotor (modeled as a rigid disk) considering gyroscopic effects is:
$$ \mathbf{M}_d \ddot{\mathbf{q}}_d + \Omega \mathbf{G}_d \dot{\mathbf{q}}_d = \mathbf{F}_d $$
where $\mathbf{q}_d = [x, y, z, \theta_x, \theta_y, \theta_z]^T$ is the displacement vector, $\mathbf{M}_d$ is the mass matrix, $\mathbf{G}_d$ is the gyroscopic matrix, $\Omega$ is the rotational speed, and $\mathbf{F}_d$ is the force vector. The shaft is modeled using Timoshenko beam elements, leading to its dynamic equation:
$$ \mathbf{M}_s \ddot{\mathbf{q}}_s + \Omega \mathbf{G}_s \dot{\mathbf{q}}_s + \mathbf{K}_s \mathbf{q}_s = \mathbf{0} $$
where $\mathbf{M}_s$, $\mathbf{G}_s$, and $\mathbf{K}_s$ are the shaft’s mass, gyroscopic, and stiffness matrices, respectively. Bearings are represented by linear spring-damper elements at specific nodes, with stiffness matrix $\mathbf{K}_B$ and damping matrix $\mathbf{C}_B$. The gear mesh interaction couples the two gear rotors. The meshing force is modeled as a nonlinear spring-damper acting along the line of action, considering time-varying mesh stiffness $k_m(t)$, static transmission error excitation $e(t)$, and backlash $b$. The mesh force can be linearized for the dynamic model, contributing to the system’s stiffness and damping matrices via a projection vector $\mathbf{V}$ aligned with the line of action. The final system-level equation of motion is assembled as:
$$ \mathbf{M}\ddot{\mathbf{q}} + (\mathbf{C} + \Omega \mathbf{G})\dot{\mathbf{q}} + \mathbf{K}\mathbf{q} + \mathbf{f}_{nl}(\mathbf{q}) = \mathbf{Q}(t) $$
where $\mathbf{M}$, $\mathbf{C}$, $\mathbf{G}$, and $\mathbf{K}$ are the global mass, damping, gyroscopic, and stiffness matrices; $\mathbf{q}$ is the global displacement vector; $\mathbf{f}_{nl}(\mathbf{q})$ is a nonlinear function representing backlash; and $\mathbf{Q}(t)$ includes external loads and excitation from static transmission error. The dynamic transmission error (DTE) is then extracted from the solution of this equation. This modeling approach provides a high-fidelity simulation environment crucial for evaluating the impact of different spur gear modification parameters on system dynamics.
The static characteristics, namely the loaded static transmission error (STE) and the maximum contact stress, are calculated using a detailed finite element analysis of the gear pair contact. A three-dimensional finite element model of the spur gears is created, incorporating the modified tooth geometry. A nonlinear contact analysis is performed under the specified load to obtain the STE curve over a mesh cycle and the distribution of contact stress on the tooth flanks. The peak-to-peak value of the STE and the maximum von Mises or Hertzian contact stress are key performance indicators. The time-varying mesh stiffness $k_m(t)$ is also derived from this static analysis by relating the applied torque to the resulting angular deflection.
The optimization problem for spur gear compound modification is defined with four primary design variables:
1. Profile modification amount, $y_1 = C_{\alpha}$.
2. Profile modification height (normalized length), $y_2 = L_{Ca}/m_n$, where $m_n$ is the normal module.
3. Profile modification power exponent, $y_3 = b$.
4. Lead crowning amount, $y_4 = C_{\beta}$.
The objectives are to minimize both the peak-to-peak value of the dynamic transmission error (DTEpp) and the maximum contact stress ($\sigma_{max}$). These are conflicting objectives; therefore, a weighted sum approach is used to formulate a single objective function. Let $G_1$ represent DTEpp and $G_2$ represent $\sigma_{max}$. The normalized objective functions are:
$$ f_1 = \frac{G_1}{G_{1,0}}, \quad f_2 = \frac{G_2}{G_{2,0}} $$
where $G_{1,0}$ and $G_{2,0}$ are the DTEpp and $\sigma_{max}$ for the unmodified spur gear pair, respectively. The overall optimization problem is then:
$$ \min_{\mathbf{y}} F(\mathbf{y}) = w_1 f_1(\mathbf{y}) + w_2 f_2(\mathbf{y}) $$
subject to:
$$ y_{1,min} \leq y_1 \leq y_{1,max} $$
$$ y_{2,min} \leq y_2 \leq y_{2,max} $$
$$ y_{3,min} \leq y_3 \leq y_{3,max} $$
$$ y_{4,min} \leq y_4 \leq y_{4,max} $$
Here, $\mathbf{y} = [y_1, y_2, y_3, y_4]^T$, and $w_1$, $w_2$ are weighting factors (both set to 0.5 for equal importance). The bounds for each variable are determined based on empirical formulas and practical experience. For instance, initial estimates for $C_{\alpha}$ and $C_{\beta}$ can be derived from ISO standards considering load-induced deflections and manufacturing tolerances.
Evaluating the objective function $F(\mathbf{y})$ for a given set of modification parameters is computationally expensive. It requires running both the static FE contact analysis and the dynamic simulation using the finite element node method. To enable efficient global optimization, the Kriging surrogate modeling technique is employed. Kriging is a powerful interpolation method that constructs an approximate model of the expensive objective function based on a limited set of sample points. The Kriging model assumes the response $y(\mathbf{x})$ is composed of a global trend model and a localized deviation:
$$ y(\mathbf{x}) = \mathbf{f}(\mathbf{x})^T \boldsymbol{\beta} + Z(\mathbf{x}) $$
where $\mathbf{f}(\mathbf{x})^T \boldsymbol{\beta}$ is a regression model (often a polynomial), and $Z(\mathbf{x})$ is a Gaussian stochastic process with zero mean, variance $\sigma^2$, and a spatial correlation function. The correlation function governs how the deviation at one point influences another, typically based on the distance between points. The model is “trained” using a Design of Experiments (DoE) to select sample points within the variable bounds. Once trained, the Kriging model can predict the objective function value at any new point almost instantly, along with an estimate of prediction error.
The optimization workflow for spur gear compound modification is as follows:
1. Define the design variable ranges for the spur gear modification.
2. Generate an initial set of sample points using a space-filling design (e.g., Latin Hypercube Sampling).
3. For each sample point:
a. Generate the 3D CAD model of the spur gears with the specified compound modification parameters.
b. Perform the FE contact analysis to obtain STE, mesh stiffness, and maximum contact stress.
c. Run the finite element node dynamic simulation using the extracted mesh stiffness and STE as input to compute DTEpp.
d. Calculate the objective function value $F$.
4. Construct the Kriging surrogate model based on the sample data.
5. Use an efficient global optimization (EGO) algorithm, which leverages the Kriging model’s prediction and uncertainty, to find the point that maximizes the Expected Improvement (EI) criterion. This point is then evaluated using the full FE/dynamic analysis, added to the sample set, and the Kriging model is updated.
6. Iterate until a convergence criterion is met (e.g., minimal improvement or maximum iterations).
To demonstrate the effectiveness of this methodology, a case study is performed on a pair of spur gears with the following parameters, which are typical in industrial applications:
| Parameter | Pinion (Z1) | Gear (Z2) |
|---|---|---|
| Normal Module, $m_n$ (mm) | 3 | 3 |
| Number of Teeth | 29 | 49 |
| Normal Pressure Angle, $\alpha_n$ (°) | 25 | 25 |
| Face Width, $b$ (mm) | 20 | 20 |
| Power (kW) | 14.8125 | 14.8125 |
| Speed (rpm) | 1000 | 591.84 |
The gear accuracy grade is 6. Additional system parameters include shaft dimensions (length 10 mm, outer diameter 30 mm, solid), bearing stiffness values (e.g., $k_{xx} = 1.49 \times 10^8$ N/m, $k_{yy} = 1.92 \times 10^8$ N/m), and damping. The variable bounds are set based on ISO-derived empirical values: $y_1 \in [5.65, 22.61]$ μm, $y_2 \in [0.71, 1.43]$, $y_3 \in [1.0, 2.0]$, $y_4 \in [4.17, 16.70]$ μm. An initial sample set of 81 points (3 levels per variable) is generated, and the Kriging-based EGO process is run.
The accuracy of the Kriging model is first verified. Five random validation points within the design space are selected. The objective function is predicted by the Kriging model and compared against the full simulation result. The relative errors are all below 1%, confirming the model’s fidelity.
| Validation Point | $y_1$ (μm) | $y_2$ | $y_3$ | $y_4$ (μm) | Predicted $F$ | Actual $F$ | Relative Error (%) |
|---|---|---|---|---|---|---|---|
| 1 | 19.47 | 0.827 | 1.656 | 13.02 | 0.8372 | 0.8365 | 0.083 |
| 2 | 21.01 | 1.407 | 1.036 | 4.57 | 0.5748 | 0.5796 | 0.839 |
| 3 | 7.81 | 1.397 | 1.849 | 7.64 | 0.7891 | 0.7899 | 0.109 |
| 4 | 21.14 | 1.061 | 1.934 | 4.75 | 0.7920 | 0.7942 | 0.287 |
| 5 | 16.37 | 1.285 | 1.679 | 5.39 | 0.7120 | 0.7071 | 0.687 |
After convergence of the optimization routine, the optimal compound modification parameters for the spur gear pair are identified. The performance of the optimized design is compared against the unmodified baseline.
| Design | $y_1$ (μm) | $y_2$ | $y_3$ | $y_4$ (μm) | DTEpp (μm) | $\sigma_{max}$ (MPa) | $F$ |
|---|---|---|---|---|---|---|---|
| Unmodified Spur Gears | 0 | 0 | 0 | 0 | 28.1627 | 1344.55 | 1.0000 |
| Optimized Spur Gears | 16.74 | 1.40 | 1.00 | 8.04 | 6.7642 | 1055.67 | 0.5126 |
The results are striking. The optimized spur gears exhibit a 75.98% reduction in the peak-to-peak dynamic transmission error (from 28.16 μm to 6.76 μm) and a 21.48% reduction in maximum contact stress (from 1344.6 MPa to 1055.7 MPa). The time-domain response and frequency spectrum of the DTE further illustrate the improvement. The unmodified spur gear pair shows significant vibration energy at the mesh frequency (2nd harmonic of the rotational speed in this case) with an amplitude of 11.72 μm. In contrast, the optimized spur gear pair demonstrates a drastically reduced mesh frequency amplitude of only 2.70 μm, indicating a much smoother and quieter operation. The compound modification effectively reshapes the mesh stiffness and STE excitation, reducing the dynamic load and improving the load distribution across the tooth face.
This case study underscores the power of the proposed integrated methodology. The Kriging surrogate model efficiently navigated the four-dimensional design space, avoiding the need for thousands of expensive full simulations. The finite element node dynamic model provided a realistic assessment of system dynamics, crucial for capturing the true impact of modifications on vibration. The final design represents a balanced trade-off between dynamic performance and contact stress for these spur gears.
The proposed novel design methodology for compound modification of spur gears successfully integrates advanced numerical techniques to address a complex multi-objective optimization problem. By combining finite element analysis for static contact evaluation, a finite element node-based dynamic model that incorporates shaft flexibility, bearing properties, and gyroscopic effects, and the efficient global optimization capability of Kriging surrogate modeling, this approach provides a robust and practical tool for gear designers. The application to a realistic spur gear pair demonstrated substantial performance gains: a reduction of over 75% in dynamic transmission error fluctuation and over 21% in maximum contact stress. These improvements directly translate to lower noise, reduced vibration, and enhanced durability of gear transmissions. The methodology is general and can be adapted to other types of spur gears or even helical gears with appropriate modifications to the geometric and dynamic models. It offers a significant step forward from rule-of-thumb or single-objective modification strategies, enabling the design of high-performance, reliable spur gear systems through systematic, simulation-driven optimization.