The rapid development of the new energy vehicle industry, propelled by the global energy crisis and tightening environmental regulations, has fundamentally transformed the automotive landscape. In recent years, the market share of electric cars has climbed steadily, driven by technological breakthroughs and supportive policies. Unlike traditional internal combustion engine vehicles, electric cars face a distinct set of challenges regarding noise, vibration, and harshness (NVH). The high-speed, high-torque characteristics of the drive motor, combined with the absence of engine noise masking, expose the vibrations and noises of other components, particularly those of the powertrain. As a key component of electric cars, the reducer’s NVH performance directly influences the overall acoustic comfort and driving quality of the vehicle. Therefore, conducting an in-depth investigation into the NVH performance of electric car reducers is of paramount importance for enhancing market competitiveness and advancing the electric vehicle industry.
My research focuses on a reducer for electric cars in the development stage. The goal is to systematically analyze and optimize its NVH performance by integrating multi-physics field modeling, simulation analysis, and experimental validation to address the specific challenges of high-speed electric drivetrains. This study developed a comprehensive methodology to analyze and resolve the complex issue of gear whine and structural vibration, applying a robust “modeling-simulation-experiment-optimization” workflow.
1. Dynamic Modeling and Foundational Analysis of the Reducer System
The research began by constructing a detailed dynamic model of the reducer, which is the foundation for all subsequent analyses. I developed a multi-degree-of-freedom dynamic model using finite element and dynamic modeling methods. The model encompasses all critical components of the reducer system: the housing, elastic shafts, gear meshing units, and bearings. To manage the model’s complexity, I used discretization and dynamic condensation techniques to derive the system’s mass, stiffness, and damping matrices. These component matrices were then used to build a complete reducer system model by integrating the HyperMesh and Romax software platforms.
The system’s governing dynamic equation, which forms the basis of the analysis, is expressed as:
$$ \mathbf{M} \ddot{\mathbf{q}}(t) + \mathbf{C} \dot{\mathbf{q}}(t) + \mathbf{K} \mathbf{q}(t) = \mathbf{F}(t) $$
where \(\mathbf{M}\), \(\mathbf{C}\), and \(\mathbf{K}\) are the global mass, damping, and stiffness matrices of the system, respectively. \(\mathbf{q}(t)\) is the displacement vector, and \(\mathbf{F}(t)\) is the external excitation force vector. For different components, this fundamental equation was adapted and expanded to capture their unique physical behaviors, as detailed in the subsequent analysis.
For model validation and structural reliability, I performed a static strength analysis on the reducer housing under extreme driving conditions. The finite element model of the housing, which was carefully meshed with elements sized at approximately 2mm, is detailed in Table 1 alongside the differential housing. The applied loads, derived from the transmission system’s bearings, and the fixed constraints at the mounting points simulated real-world installation.
| Parameter | Value |
|---|---|
| Rated Torque (Nm) | 140 |
| Peak Torque (Nm) | 320 |
| Rated Power (kW) | 60 |
| Peak Power (kW) | 130 |
| Maximum Speed (rpm) | 10000 |
The material properties assigned to the housing and differential in the finite element model, which are critical for accurate stress and modal analysis, were based on supplier information and are presented in Table 2.
| Component | Material | Elastic Modulus (MPa) | Density (kg/m³) | Poisson’s Ratio |
|---|---|---|---|---|
| Reducer Housing | A380 Aluminum | 70000 | 2700 | 0.33 |
| Differential Housing | QT500 Cast Iron | 169000 | 7100 | 0.275 |
Through HyperMesh and the OptiStruct solver, I evaluated the structural integrity at the limits of operation. Under the forward-driving extreme condition (320 Nm of peak torque), the maximal stress recorded was 110.889 MPa. Under the reverse-driving condition, the stress peaked at 98.861 MPa. Both values are well below the material’s permissible stress of 160 MPa. The displacement analysis showed a maximum deformation of 0.082 mm, which is substantially less than the 2.445 mm limit allowed by industry standards for this vehicle’s wheelbase. These results affirm the structural soundness of the reducer housing under rigorous operational demands, ensuring it is a reliable platform for the detailed NVH investigation that follows.
The material properties for the key components ensure accurate modeling across different simulation environments.
| Component | Material | Elastic Modulus (MPa) | Density (kg/m³) | Poisson’s Ratio |
|---|---|---|---|---|
| Drive Shaft | 45 Steel | 206000 | 7850 | 0.29 |
| Gear Pair | 20CrMnTi | 210000 | 7870 | 0.30 |
2. Modal Analysis, Vibrational Response, and Excitation Source Identification
The study of the reducer’s dynamic behavior began with a modal analysis of its housing. Identifying the natural frequencies and mode shapes is crucial to ensure no resonance occurs with the excitation frequencies of the gear system. I conducted free modal simulations using the finite element model and complemented this with an experimental modal analysis using the hammer impact method.
The fundamental equation for the undamped free vibration modal analysis is derived by setting the external force and damping to zero:
$$ \mathbf{M} \ddot{\mathbf{x}}(t) + \mathbf{K} \mathbf{x}(t) = 0 $$
Solving this eigenproblem provides the natural frequencies, \(\omega_0\), and mode shapes of the housing. The results, comparing the simulated and experimentally measured natural frequencies for the first six modes, show a strong correlation, as displayed in Table 4. The maximum relative error was only 2.67%, conclusively verifying the accuracy of the finite element model. This high correlation validated the model for use in further dynamic response predictions.
| Mode Order | Simulation Frequency (Hz) | Experimental Frequency (Hz) | Relative Error (%) |
|---|---|---|---|
| 1 | 1236.7 | 1267.8 | 2.45 |
| 2 | 1301.4 | 1291.9 | 0.74 |
| 3 | 1539.7 | 1559.8 | 1.29 |
| 4 | 1846.3 | 1798.3 | 2.67 |
| 5 | 1891.6 | 1884.2 | 0.39 |
| 6 | 1941.5 | 1965.0 | 1.20 |
The verification of the modal test data was performed using the Modal Assurance Criterion (MAC). The MAC matrix, which indicates the correlation between mode shape vectors, is a standard tool for comparing experimental and analytical modal data. A good correlation between the two data sets is indicated by high values (close to unity) along the matrix diagonal and low values on the off-diagonal elements, confirming that the experimentally extracted modes are independent and accurately represent the structure’s dynamic characteristics.
Having confirmed its accuracy, I used the model to analyze the reducer’s dynamic response under excitation. To identify the worst-case condition, I performed a gear static transmission error analysis at various input torques. The key excitation source for NVH, gear transmission error, is a measure of non-uniform motion in the gear pair. The analysis revealed a non-monotonic relationship between the transmission error peak-to-peak value and the input torque.
The examination of gear transmission error showed a significant correlation with input torque, rising from 2.63 μm at 10 Nm to 9.00 μm at 140 Nm. Interestingly, the peak-to-peak values of the error initially decreased, then increased with torque, reaching their maximum at an input torque of 70 Nm. From 10 Nm to approximately 70 Nm, the gear transmission error peaked. This initial decrease is because the load-induced deformation at lower torques counteracts some of the geometric errors from manufacturing, optimizing contact. Beyond 100 Nm, the deformation caused by the increasing load becomes the dominant source of transmission error.
I then selected the first-order harmonic at this 70 Nm operating point as the primary excitation source for the dynamic simulations. The response analysis of the reducer housing, detailed in Figure 1, employed a “gear transmission error – gear meshing force – housing vibration” workflow.
Under the 70 Nm / 5800 rpm operating condition, the vibration response analysis of the reducer housing produced several key findings. The RMS velocity spectrum of the housing surface showed three distinct vibration peaks centered around 816 Hz, 2640 Hz, and 3200 Hz. By comparing these frequencies with the system’s calculated natural frequencies from 324.8 Hz to 3899.3 Hz, I identified the specific modes contributing to these peaks. The system natural frequencies associated with these peaks are presented in Table 5.
| Vibration Peak (approx.) | Dominant System Natural Frequencies (Hz) | Associated Mode Orders |
|---|---|---|
| 816 Hz | 808.4 | 7 |
| 2640 Hz | 2451.4, 2594.9, 2723.6 | 22, 23, 24 |
| 3200 Hz | 3237.4 | 32 |
To pinpoint the source of the housing vibration, I performed a modal contribution analysis. This analysis clarifies which natural mode is the primary driver of vibration at a given frequency. The results were conclusive, showing dominant modes (contributing over 91%) for each primary vibration peak:
- At ~816 Hz, the 7th system mode was the primary contributor to the vibration at the input shaft right bearing and the output shaft left bearing.
- At ~2640 Hz, the vibration was mainly driven by the 23rd and 24th modes, affecting the intermediate shaft bearings.
- At ~3200 Hz, the 32nd mode was responsible for the vibration transmitted through the input shaft left and intermediate shaft left bearings.
Further strengthening these findings, the dynamic contact load and modal flexibility analyses showed distinct peaks at these same frequencies (816, 2640, and 3200 Hz), indicating a strong correlation and confirming that the identified modes were susceptible to excitation. It is important to note that while peaks existed, the magnitude of the vibration acceleration under this condition (maximum around 9.8 m/s²) was below typical limits for acceptable NVH performance, indicating the system was within design requirements but had potential for improvement.
3. Radiated Noise Analysis and Experimental Verification
Building on the established correlation between structural vibration and acoustic radiation, I proceeded to analyze the reducer’s noise output. The noise from the reducer is primarily structure-borne, originating from the internal gear excitations. The vibrational energy is transmitted through the bearings to the housing, whose vibrating surfaces then couple with the surrounding air to generate sound waves. I established the relationship between vibration and radiated noise. The sound pressure level (\(L_p\)) was calculated using the following formula:
$$ L_p = 20 \log{\frac{p_e}{p_r}} $$
where \(p_e\) is the measured effective sound pressure and \(p_r\) is the reference sound pressure (20 μPa in air). To correlate sound power with the acoustic environment, I used the following expression for the sound power level (\(L_W\)):
$$ L_W = 10 \log{\frac{W}{W_0}} $$
where \(W\) is the sound power and \(W_0\) is the reference sound power of 10⁻¹² watts.
To assess the acoustic performance, I constructed a boundary element model of the reducer housing’s outer surface, ensuring a fully sealed geometry by filling all bolt holes and vents. For an acoustic analysis, mesh size is determined by the highest frequency of interest (4000 Hz) and the speed of sound in air, using the rule of six elements per wavelength:
$$ L = \frac{c}{6 \times f_{\text{max}}} = \frac{343000 \text{ mm/s}}{6 \times 4000 \text{ Hz}} \approx 14 \text{ mm} $$
Four microphone field points were established at a distance of 1 meter from the housing’s geometric center, positioned in the front, rear, left, and right directions to capture the noise radiation directivity. I used an A-weighted sound pressure level to reflect the human auditory system’s sensitivity. The simulation of the radiated noise showed that the main peaks in the sound pressure level (SPL) spectrum occurred at frequencies of approximately 810 Hz, 2600 Hz, and 3230 Hz. These frequencies were in close agreement with the peaks found in the structural vibration analysis (816 Hz, 2640 Hz, 3200 Hz), confirming that the vibration is the primary source of the radiated noise.
To validate my simulation model, I performed a noise benchmark test in an NVH semi-anechoic chamber using a T-type test bench. The noise testing equipment and specific parameters are listed in Table 6.
| Equipment | Model | Sensitivity |
|---|---|---|
| PCB Microphone | 378B02 | 53 mV/Pa |
| Photoelectric Speed Sensor | CYT70-N | – |
By comparing the simulation results with the experimental data at the top and rear microphone positions, I observed compelling agreement. The frequency spectrum of the radiated noise, particularly in identifying the primary broadband peaks, matched well between simulation and test. This agreement validated the accuracy of my analysis chain. It provided experimental confirmation of my simulation model and, more importantly, confirmed the physical relationship between the gear excitations, the structural vibrations of the housing, and the resulting acoustic field. The slight variations in the amplitude of the peaks can be attributed to differences between the idealizations in the simulation and the complexities of the physical test, such as the load motor’s torque fluctuations.
4. Optimization of NVH Performance via Gear Micro-Modification & NSGA-II
The source of the vibration and noise peaks identified in the previous stages can be traced to the gear tooth meshing process. As gear teeth engage and disengage, they generate internal dynamic excitation forces. These forces arise from changes in meshing stiffness, transmission errors due to manufacturing imperfections and load-induced tooth deflection, and the impact forces generated as teeth enter and leave the mesh. To mitigate these internal excitations and thereby reduce the NVH response of the reducer, I focused my optimization strategy on the gear tooth micro-geometry. The objective was to modify the tooth surfaces so that the load is distributed more evenly and the dynamic transmission error is minimized, leading to smoother and quieter operation.
The optimization methodology employed a multi-objective genetic algorithm, specifically the Non-dominated Sorting Genetic Algorithm II (NSGA-II). The design variables were the micro-modification parameters for the high-speed first-stage gear pair: helix Crowning, helix slope, involute Crowning, and involute slope. The optimization objectives were to minimize both the gear transmission error peak-to-peak value and the maximum normal load per unit length of the tooth surface. The constraints for the gear pair in the electric cars were ensuring adequate gear strength (safety factor for contact and bending fatigue). The design variable ranges were carefully chosen based on industry best practices.
The optimization process is a multi-step process that includes:
- Randomly generating an initial population of 50 potential design parameter sets (chromosomes).
- Simulating the gear contact for each parameter set under the relevant load case (70 Nm and 5800 rpm).
- Evaluating the performance of each design based on the two optimization objectives.
- Applying a penalty strategy to designs that do not meet the strength constraints.
- Ranking the resulting designs using the NSGA-II algorithm’s fast non-dominated sorting and crowding distance procedures.
- Creating a new generation of design solutions through genetic operations (selection, crossover, and mutation).
- Repeating the process for 20 generations, resulting in a total of 1000 potential solutions.
From the resulting Pareto front of optimal designs, which represented the trade-off between the competing objectives of minimal transmission error and minimal tooth loading, I selected a final design solution. The selected parameters are shown in Table 7.
| Modification Parameter | Active Gear | Passive Gear |
|---|---|---|
| Helix Crowning (μm) | 0.06 | 0.13 |
| Helix Slope (μm) | -13.78 | 16.45 |
| Involute Crowning (μm) | 1.24 | -2.74 |
| Involute Slope (μm) | 2.37 | 4.53 |
A comparative analysis of the numerically predicted NVH performance for the optimized gear design versus the baseline design is presented in Table 8. The contact pattern analysis showed that the optimized tooth surface has a consistent, elliptical contact pattern centered on the tooth flank, indicating a more uniform load distribution near the middle of the tooth, reducing edge-loading and stress concentration. This optimization resulted in substantially lower housing vibration and radiated noise.
| Performance Metric | Frequency Point | Value | Reduction (%) | |
|---|---|---|---|---|
| Before | After | |||
| Housing Vibration RMS (m/s) | ~816 Hz | 4.81 × 10⁻⁹ | 4.09 × 10⁻⁹ | 15.0% |
| ~2640 Hz | 9.57 × 10⁻⁹ | 7.35 × 10⁻⁹ | 23.2% | |
| ~3200 Hz | 5.05 × 10⁻⁹ | 3.40 × 10⁻⁹ | 32.7% | |
| Radiated Noise SPL (dBA) | ~810 Hz | 54.7 | 51.0 | -3.7 dB |
| ~2600 Hz | 57.8 | 53.7 | -4.1 dB | |
| ~3230 Hz | 58.5 | 54.0 | -4.5 dB | |
The results presented in Table 8 and the accompanying analysis conclusively show that the gear micro-modification optimization strategy is highly effective. The strongest attenuation was observed at higher frequencies (3200 Hz), where the vibration RMS was reduced by nearly a third. This significant reduction in housing vibration led to corresponding reductions in the overall radiated noise, with a peak reduction of 4.5 dB(A) at 3230 Hz, and an average reduction of 3.9 dB(A) across the peaks. The implementation of this optimized gear design not only suppressed the key vibration peaks but also substantially improved the acoustic comfort of the electric car’s drivetrain, demonstrating a powerful methodology for NVH refinement.
In summary, through the comprehensive methodology of theoretical modeling, simulation, and targeted optimization, I successfully identified the root causes of a specific electric car reducer’s NVH issues and applied a sophisticated optimization algorithm to mitigate them. The entire workflow, from dynamic model creation and experimental validation to acoustic prediction and genetic algorithm-based optimization, provided a robust and effective strategy for enhancing the NVH performance of critical components in modern electric cars. This research underscores the importance and effectiveness of dynamic excitation source control for developing quieter and more refined electric vehicles.