The Sound of Silence as AI Rewrites EV NVH Engineering

Step into a modern electric vehicle, and one thing becomes immediately clear: the absence of engine noise changes what passengers hear. Sounds that were once masked by combustion engines, tire roar on coarse asphalt, wind rush around mirrors, HVAC fan noise, cooling pumps, and the high-frequency whine of an electric drive unit now stand out more clearly in the cabin. This is why cabin comfort has become a defining measure of perceived vehicle quality, and why noise, vibration, and harshness (NVH) engineering is moving from late-stage refinement to a core design priority.

What begins as a noticeable shift in the passenger experience quickly becomes an engineering-scale issue. EV powertrains introduce multiphysics NVH problems that span electromagnetics, structural vibration, acoustics, thermal systems, and vehicle dynamics. As these coupled domains multiply the number of possible design trade-offs, high-fidelity simulation remains essential, but traditional workflows can struggle to evaluate every alternative within practical timelines.

The multiphysics chain for e-motor NVH, mapping internal electromagnetic excitation forces to structural housing vibrations and downstream acoustic radiation

The real opportunity is to extend physics-based simulation with artificial intelligence (AI), combining high-fidelity multiphysics tools such as ACTRAN, MSC Nastran, and Cradle CFD with AI-driven surrogate modeling and optimization capabilities from ODYSSEE (ODC). Together, these technologies help engineers move from computationally expensive analysis toward real-time design exploration while keeping the workflow grounded in engineering physics.

The Growing Complexity of Modern NVH Programs

Modern NVH engineering is fundamentally a coupled multiphysics challenge. Because these domains are tightly linked, engineers rarely solve a single isolated problem: a change to improve motor efficiency, reduce mass, stiffen a bracket, or update trim can shift vibration paths and acoustic response elsewhere in the vehicle.

Motor whine analysis in an EV illustrates the challenge. Electromagnetic forces in the motor can excite the stator, housing, mounts, and surrounding structure, which then radiate into the cabin as sound. A small change in motor geometry, inverter switching, housing stiffness, or mount design may improve one metric while creating a tonal peak audible to passengers.

At the system level, cross-domain interactions become harder to manage. Today’s computing resources can solve high-fidelity NVH models, but running hundreds or thousands of design iterations remains too costly and time-consuming.

Why EVs Are Raising the Bar for Acoustic Engineering

The shift from combustion engines to electric powertrains has changed what engineers must control. With engine and exhaust noise reduced, sources such as inverter whine, gear noise, road noise, and wind noise become more noticeable.

Details such as tire pattern, mirror shape, cooling-pump behavior, or a lightly damped panel can now influence whether the cabin feels refined or fatiguing. Active road-noise cancellation systems show how seriously automakers treat low-frequency road noise and structure-borne vibration as part of the cabin experience.

OEMs, therefore, need to optimize acoustic treatments and sound packages across many combinations of absorber materials, barrier layers, foam thicknesses, mounting locations, and packaging constraints—an ideal fit for AI-assisted engineering workflows.

ACTRAN: Capturing Physics That AI Can Learn From

High-fidelity physics simulation is the foundation of trustworthy engineering AI. ACTRAN supports acoustic, vibro-acoustic, and aero-acoustic simulation, including interior and exterior noise, sound quality, structural vibration, poroelastic materials, fluid-structure interaction, and frequency- and time-domain behavior. By capturing complex sound propagation and NVH responses, ACTRAN generates physics-rich data that helps AI models learn how design choices affect acoustic performance.

For a firewall case study, ACTRAN models the acoustic barrier between the engine compartment and passenger cabin. Engineers define acoustic treatment layers by material, sequence, thickness, and placement, while ACTRAN predicts performance using metrics such as panel transmission loss. The firewall must reduce airborne noise, control structure-borne vibration paths, and still meet packaging and weight constraints.

Panel transmission loss measures how effectively an assembly blocks noise across frequencies; higher values indicate stronger insulation and better cabin comfort.

The challenge here is identifying the best candidate from many technically valid options.

For a single firewall design, engineers may vary material sequence, layer thickness, acoustic-layer count, material combinations, and placement. Simulating every configuration is too slow for real-world design timelines.

Building Surrogate Models with ODYSSEE

This is where Cadence ODYSSEE (ODC) extends the workflow.

ODYSSEE is an AI and machine-learning platform for engineering applications. It learns relationships between design variables and responses using data from CAE tools, tests, sensors, CAD models, and other engineering sources.

In the firewall case study, engineers encode acoustic materials numerically, define thickness limits and constraints, generate candidate designs through DOE, run ACTRAN simulations, and import the results into ODYSSEE for model training.

The trained model becomes a surrogate representation of the underlying physics. Once validated, it can predict performance almost instantly, allowing engineers to test “what if” scenarios without launching a full finite element analysis for every design change.

AI as the Integration Layer for Multiphysics Optimization

One of AI’s most important roles is bridging disciplines.

A machine-learning model can combine variables from vehicle dynamics, structural mechanics, acoustics, and electromagnetics, even though these domains follow different governing equations.

This helps engineers explore trade-offs that would otherwise require long chains of interacting simulations. A lighter acoustic package may reduce cost and mass but lower transmission loss at certain frequencies; a stiffer structure may suppress one vibration mode while shifting energy into another. Instead of optimizing one subsystem in isolation, teams can evaluate integrated system behavior.

Making AI Trustworthy for NVH Engineering

Engineering organizations need more than prediction speed; they need trust. A fast model is not useful unless engineers know where it is reliable, which variables matter most, and whether the training data covers the design space. ODYSSEE supports that confidence with explainable AI capabilities, including training-data heatmaps, sensitivity analyses, correlation studies, leave-one-out validation, and adaptive DOE recommendations.

These capabilities help identify regions where additional training data may improve accuracy and reveal which design parameters have the greatest impact on performance at specific frequencies.

This keeps engineering expertise central to the ML workflow. Domain knowledge still guides which variables to include, which constraints matter, where more simulations are needed, and how to interpret the final recommendations, especially in NVH, where objective metrics must align with human perception.

Real-Time Design Exploration and Optimization

Once validated, ODYSSEE uses the surrogate model to optimize designs within the ML environment, without new CAD updates, meshing, or solver runs. In an acoustic-trim study using ODYSSEE, material sequence and thickness were optimized, design constraints were enforced, and multiple optimization iterations were executed automatically. The workflow completed 56 optimization runs in just 23 seconds.

The optimized solution improved the layer arrangement while satisfying physical constraints. More importantly, it showed how teams can move from evaluation to exploration, testing many alternatives, comparing trade-offs, and converging on stronger candidates earlier in development.

Quieter EVs Need Smarter NVH Exploration

As EVs become quieter, the acoustic margin for error gets smaller. Cabin comfort now depends on how quickly teams can understand coupled NVH behavior, compare more design alternatives, and make decisions that are both data-driven and physics-grounded. That is the real value of combining ACTRAN with ODYSSEE: high-fidelity simulation captures the acoustic and vibro-acoustic response, while AI-powered surrogate modeling turns that knowledge into faster exploration and optimization.

Don’t miss our new e-book, Engineer’s Guide to Modern NVH Simulation

Learn how modern NVH teams are using MSC Nastran, Actran, Romax, Digimat, and ODYSSEE to diagnose noise sources, model coupled structural-acoustic behavior, optimize trim and materials, accelerate high-frequency workflows, and explore more design alternatives with AI- and machine learning-driven reduced-order models. 

Also get access to practical methods, tool-selection guidance, and real-world results from Audi, Volvo Cars, Volvo Group, Hyundai, and Daihatsu, all aimed at helping engineers design quieter, more refined vehicles faster.