New Method Improves Accuracy in EV Vibration Analysis
In the rapidly evolving world of electric vehicle (EV) engineering, one of the most persistent challenges lies in managing noise, vibration, and harshness (NVH). As EVs eliminate the masking effect of traditional internal combustion engines, subtle vibrations from the powertrain become more perceptible, demanding higher precision in design and analysis. A recent breakthrough by a team of researchers from South China University of Technology and Guangzhou City University of Technology offers a significant leap forward in how engineers model and predict the dynamic behavior of EV mounting systems. Their work, published in the Journal of Hunan University (Natural Sciences), introduces a novel approach that more accurately captures the complex interplay of uncertain parameters in suspension design, potentially reshaping how automakers approach NVH optimization.
The study, led by Dr. Lü Hui, Associate Professor at South China University of Technology’s School of Mechanical and Automotive Engineering, in collaboration with colleagues Liao Zeyun, Li Changyu, Shangguan Wenbin, and Xiao Guoquan, focuses on the powertrain mounting system (PMS)—a critical component responsible for isolating vibrations generated by the electric motor and drivetrain from the vehicle chassis. Traditionally, engineers have relied on simplified models to predict the system’s inherent characteristics, such as natural frequencies and decoupling rates. These models often treat uncertain parameters—like the stiffness of rubber mounts—as independent variables with fixed ranges. However, in real-world manufacturing and operation, these parameters are not only uncertain but also correlated. For instance, the three directional stiffness values (X, Y, Z) of a single rubber mount may vary together due to material properties and production tolerances, while being independent of those in other mounts. Ignoring these correlations can lead to overly conservative or inaccurate predictions, ultimately affecting ride comfort and durability.
Existing methods have attempted to address uncertainty using probabilistic models or interval analysis. Probabilistic approaches require extensive statistical data, which is often unavailable in early design stages. Interval methods, while simpler, assume independence among parameters and can result in large, unrealistic uncertainty domains that include regions with no physical basis. More advanced models, such as multidimensional ellipsoids or parallel hexahedrons, have been proposed to account for parameter correlations, but they assume regular geometric boundaries. In practice, real-world parameter samples often form irregular clusters that do not conform to these idealized shapes, leading to inaccuracies in the predicted response ranges.
To overcome these limitations, the research team introduced a new modeling framework based on the Polygonal Convex Set (PCS) model, combined with Principal Component Analysis (PCA) and Monte Carlo simulation. The PCS model is designed to handle the dual nature of uncertainty in PMS: some parameters are correlated within a group (e.g., the three stiffness directions of one mount), while different groups (e.g., different mounts) remain independent. By applying PCA, the method identifies the principal directions of variation in the parameter space, effectively rotating the coordinate system to align with the natural structure of the data. This transformation allows for a more compact and realistic representation of the uncertainty domain.
The core innovation lies in the intersection of two uncertainty models: the traditional interval model and the PCA-based interval model. The traditional interval model defines a hyper-rectangle that bounds all possible parameter values, but it often includes vast regions with no actual data points—so-called “empty spaces.” The PCA-based model, while more compact, can extend beyond the physical bounds of individual parameters in certain directions. By taking the intersection of these two models, the PCS approach creates a tighter, more realistic uncertainty domain that fully contains the actual sample data while excluding non-physical combinations. This results in a more accurate estimation of the system’s response, particularly for critical metrics like decoupling rate—the measure of how isolated a particular mode of vibration is from others.
To validate their method, the team applied it to a real-world case: a three-point mounting system from an electric passenger car. The vehicle’s powertrain had a mass of 70 kg, and the initial dynamic stiffness values of the left, right, and front mounts were used as baseline parameters. The researchers introduced a ±10% uncertainty in the stiffness values and generated synthetic sample data with varying degrees of correlation—weak (0.1), moderate (0.4), and strong (0.7)—to simulate different manufacturing scenarios. Using Monte Carlo simulation with one million samples, they compared the performance of the proposed PCS model against traditional interval analysis and the multidimensional parallel hexahedron (MP) model.
The results were compelling. When parameter correlation was weak, all three models produced similar results, as expected. However, as correlation increased, the differences became pronounced. The traditional interval model, unable to account for correlations, produced the widest and most conservative response ranges for decoupling rates in the X, Z, and θY (motor rotation) directions. The MP model, while slightly better, still retained significant empty regions in its uncertainty domain, leading to only modest improvements in prediction accuracy. In contrast, the PCS model demonstrated a much tighter confinement of the response, effectively narrowing the predicted range of decoupling rates as correlation increased.
For example, under strong correlation, the X-axis decoupling rate (DX) predicted by the interval model ranged from 45.67% to 81.69%. The MP model reduced this range to 49.25%–79.67%, a modest improvement. However, the PCS model achieved a significantly tighter range of 59.77%–75.75%, indicating a more precise and realistic prediction. Similar trends were observed for the Z-axis (DZ) and θY-axis (DθY) decoupling rates. The reduction in the upper bound and the increase in the lower bound reflect a more accurate capture of how correlated stiffness variations constrain the system’s dynamic behavior.
One of the most significant findings was the differential impact of correlation on the upper and lower bounds of the response. Across all directions, the lower bounds of the decoupling rates increased more dramatically with correlation than the upper bounds decreased. This suggests that parameter correlation tends to push the system toward more favorable (higher) minimum performance levels, reducing the risk of poor NVH performance. The θY direction showed the greatest sensitivity, with the lower bound increasing by nearly 13% as correlation went from zero to 0.9, highlighting the importance of considering directional effects in design.
The study also investigated which mounts had the greatest influence on system performance. By selectively introducing correlation in only one mount at a time (with a correlation coefficient of 0.6), the team found that the right and front mounts had the most significant impact. Specifically, correlation in the right mount strongly affected the upper bound of the X-axis decoupling rate, while the front mount had a major influence on the lower bounds of both the X and Z directions. This insight is crucial for design engineers: it suggests that tighter quality control on the stiffness consistency of the right and front mounts could yield greater improvements in overall NVH performance than treating all mounts equally.
The implications of this research extend beyond academic interest. In the competitive EV market, where customer expectations for quiet, smooth operation are high, even small improvements in NVH can translate into significant advantages. Automakers investing in advanced simulation tools can use the PCS method to design more robust mounting systems that perform consistently across production variations. This reduces the need for costly physical prototypes and late-stage design changes, accelerating time-to-market.
Moreover, the method’s non-probabilistic nature makes it particularly suitable for early design stages, where comprehensive statistical data may not yet be available. Engineers can use limited test data or expert judgment to define parameter ranges and correlations, then apply the PCS model to explore the design space more efficiently. This supports a more systematic approach to robust design, where uncertainty is not ignored but actively managed.
The research also opens doors for future work. While the current study focused on stiffness parameters, the PCS framework could be extended to other uncertain factors, such as mount positions, angles, or even material damping properties. Additionally, the method could be integrated into optimization routines to find mount configurations that maximize decoupling rates while minimizing sensitivity to parameter variations. As EVs continue to evolve—with heavier batteries, higher torque motors, and new architectures like skateboard platforms—the need for advanced uncertainty analysis will only grow.
From a broader perspective, this work exemplifies the shift in automotive engineering toward data-driven, model-based design. As vehicles become more complex systems of systems, traditional “rule-of-thumb” approaches are no longer sufficient. Instead, engineers must leverage advanced mathematical tools to understand and control the interactions between components. The PCS model represents a step in this direction, offering a more nuanced and realistic way to handle uncertainty.
For industry practitioners, the takeaway is clear: ignoring parameter correlations in mounting system design can lead to suboptimal performance and unnecessary over-engineering. By adopting more sophisticated uncertainty modeling techniques like the PCS approach, engineers can achieve a better balance between performance, cost, and reliability. As EV technology matures, such precision will become a key differentiator in the quest for superior driving experiences.
The automotive industry stands at a crossroads, where innovation in software, materials, and manufacturing converges to redefine what is possible. This research from South China University of Technology and Guangzhou City University of Technology is a testament to the power of interdisciplinary thinking—combining mechanical engineering, data science, and applied mathematics to solve a real-world problem. It underscores the importance of academic research in driving industrial progress, especially in emerging fields like electric mobility.
As global automakers race to electrify their fleets, the ability to deliver a premium NVH experience will be a critical factor in consumer adoption. Methods like the one developed by Lü Hui and his team provide the tools needed to meet this challenge, ensuring that the quiet promise of electric vehicles is fully realized on the road.
Lü Hui, Liao Zeyun, Li Changyu, Shangguan Wenbin, Xiao Guoquan, South China University of Technology, Guangzhou City University of Technology, Journal of Hunan University (Natural Sciences), DOI: 10.16339/j.cnki.hdxbzkb.2024185