Automotive Noise, Vibration, and Harshness (NVH): A Thematic Literature Review

Abstract

Automotive Noise, Vibration, and Harshness (NVH) has emerged as a critical interdisciplinary field influencing vehicle performance, passenger comfort, brand perception, and regulatory compliance. This thematic literature review synthesizes key research trends, methodological approaches, and technological developments shaping contemporary NVH studies. Drawing on 255 scholarly and industry sources, the review identifies five dominant themes: (1) sources and characterization of noise and vibration in internal combustion, hybrid, and electric vehicles; (2) advanced modeling and simulation techniques—including finite element analysis, statistical energy analysis, and machine learning–based prediction models; (3) materials, components, and structural optimization strategies for NVH mitigation; (4) the rapidly evolving landscape of electric and autonomous vehicle NVH; and (5) emerging active noise and vibration control technologies and data-driven diagnostics. The analysis highlights a definite shift toward holistic, data-driven, and multi-physics approaches, driven by lightweighting imperatives, widespread electrification, and increasingly stringent occupant comfort expectations. Key gaps in current research—including the need for unified evaluation metrics, real-time in-vehicle NVH monitoring, closer integration of subjective psychoacoustic perception with objective physical measurement, and validated simulation workflows for novel EV architectures—are identified and discussed. This review provides a consolidated and expanded framework for understanding contemporary NVH research directions and articulates opportunities for transformative innovation in next-generation vehicle development.

1. Introduction

Noise, vibration, and harshness (NVH) is a multidisciplinary field encompassing acoustics, structural dynamics, psychoacoustics, materials engineering, and systems modeling. As modern vehicles have become quieter due to improved powertrain refinement, electrification, and advanced body structures, customer expectations for comfort have increased dramatically. NVH has therefore evolved from a quality attribute to a primary design driver in automotive engineering. The literature shows a continuous shift from empirical troubleshooting to predictive, simulation-led development and integrated sound quality engineering [1,2,3,4,5]. At the same time, the rise of electric vehicles (EVs) has transformed NVH priorities: while powertrain noise decreases, secondary noise sources such as tire–road interaction, wind noise, and accessory systems become more dominant [6,7,8,9]. This transformation can be quantified in terms of the masking threshold shift: in a typical ICE vehicle cruising at 100 km/h, broadband combustion and drivetrain noise raises the cabin interior sound pressure level to approximately 65–72 dB(A), providing 15–25 dB(A) of masking headroom above the perceptibility thresholds of secondary sources [6,7]. In a comparable battery electric vehicle under identical conditions, the interior level drops to approximately 52–60 dB(A), reducing or eliminating this masking margin and rendering previously inaudible tonal sources (tire cavity resonance at ~200 Hz, e-motor whine at 1–8 kHz, inverter harmonics) clearly perceptible by occupants [6,7,8,9,10,11,12]. This 12–15 dB(A) reduction in broadband masking floor is the defining psychoacoustic challenge of EV NVH and motivates the shift toward tonality-calibrated evaluation metrics discussed in Section 2.2 and Section 2.3.

This review synthesizes 255 scholarly and industry sources to describe the state of the art in NVH science, measurement, modeling, and mitigation within the automotive domain. The literature search was conducted systematically across four principal databases: Scopus, Web of Science, SAE Mobilus, and Google Scholar, supplemented by targeted searches in the AIAA Digital Library and IEEE Xplore for aeroacoustics and electromechanics content respectively. Search queries were constructed using Boolean combinations of primary terms (“automotive NVH”, “vehicle noise vibration harshness”, “road noise”, “wind noise vehicle”) combined with thematic modifiers for each section (e.g., “electric vehicle noise”, “acoustic metamaterial vehicle”, “transfer path analysis”, “statistical energy analysis automotive”, “finite element NVH”). The search covered publications from 1952 (Lighthill’s aeroacoustic theory) through 2025, with no lower date boundary applied, to capture foundational works alongside current research. Sources were screened by title and abstract relevance to automotive NVH, then assessed for citation count, journal impact, and methodological rigor. Grey literature (SAE Technical Papers, OEM technical reports) was included where peer-reviewed equivalents were unavailable for specific industry-validated methods. The discussion is organized into theoretical foundations, major noise and vibration sources, measurement and testing methods, computational approaches, materials and countermeasures, and emerging trends in EV and active control technologies. The remainder of the paper is structured as follows. Section 2 covers the fundamentals of acoustic and vibrational theory. Section 3 surveys the major noise sources. Section 4 addresses vibration phenomena and harshness. Section 5 and Section 6 review measurement techniques and computational modeling approaches, respectively. Section 7 discusses materials and countermeasures. Section 8, Section 9, Section 10 and Section 11 address EV-specific NVH, active control, future research directions, and digital development workflows. Section 12 concludes.

2. Fundamentals of NVH

2.1. Acoustic and Vibrational Theory

NVH analysis relies heavily on classical acoustics and structural vibration theory. Key contributions on wave propagation, acoustic radiation, and transmission loss originate from foundational works such as Beranek’s acoustics treatises [13], Cremer’s structural–acoustic coupling models [14], and Lyon & DeJong’s statistical approaches [15]. Modal analysis, which forms the basis for identifying natural frequencies, mode shapes, and damping characteristics, is documented comprehensively in Ewins [16] and Maia & Silva [17].

The fundamental distinction between airborne and structure-borne noise is critical for diagnosing vehicle-level NVH issues. Airborne noise, governed by acoustic domain equations, transmits through leak paths, panels, and interior cavities, whereas structure-borne noise is generated via mechanical excitation transmitted through the chassis, mounts, and body structure [18,19,20,21]. A thorough understanding of energy transmission mechanisms underpins the transfer path analysis (TPA) methodologies described in Section 5.3 and the countermeasure selection strategies presented in Section 7.

2.2. Sound Quality and Psychoacoustics

As vehicles have become quieter, subjective sound quality has gained prominence alongside objective physical measurement. Psychoacoustic metrics such as loudness, sharpness, roughness, fluctuation strength, and tonality provide diagnostic insight beyond A-weighted sound pressure levels [22,23,24,25]. Zwicker’s loudness model [25], Fastl & Zwicker’s psychoacoustic framework [25], and the Sensory Pleasantness indices proposed in automotive studies [26,27] are widely employed by original equipment manufacturers (OEMs). Sound quality engineering is now integrated into early design stages, especially for EVs, where traditional engine masking noise is absent and previously inaudible tonal components become perceptually dominant [10,11,12].

The convergence of psychoacoustic evaluation with objective simulation—discussed further in Section 11 in the context of virtual acoustic prototyping—represents one of the most productive research frontiers in contemporary NVH engineering, enabling listening panel assessments of synthesized acoustic scenarios before physical prototypes exist [10,11]. A complementary and increasingly important dimension is soundscape characterization, which situates individual NVH metrics within the broader perceptual context of the acoustic environment as experienced holistically by occupants. Unlike classical psychoacoustic metrics that decompose perception into single-attribute scores, soundscape methods assess the overall character of an acoustic environment on perceptual dimensions such as eventfulness and pleasantness, as standardized in ISO 12913 [28]. Indoor perception tests applying soundscape methodology to vehicle cabin NVH have demonstrated that identical A-weighted SPL levels produce markedly different subjective responses depending on the spectral, temporal, and informational content of the sound [29]. In the EV context, soundscape approaches have been applied to assess the perceptual acceptability of motor whine and ASD synthetic sounds in isolation and in combination with road and wind noise backgrounds, revealing interactions between annoyance dimensions that single-metric analyses miss [29]. These methods offer a validated pathway for bridging objective multiphysics NVH simulation with qualitative occupant perception data, and are emerging as a complement to traditional listening panel protocols in advanced OEM sound quality programs.

2.3. Core NVH Evaluation Metrics

To facilitate consistent interpretation of the noise reduction data cited throughout this review, this section provides a consolidated reference for the principal NVH evaluation metrics employed in the reviewed literature. These metrics span three categories: physical acoustic measurements, psychoacoustic descriptors, and intelligibility indices.

Physical acoustic metrics include Sound Pressure Level (SPL), expressed in decibels (dB) as 20·log10(p/p0) where p0 = 20 μPa, which quantifies the instantaneous acoustic energy at a measurement point. A-weighted SPL [dB(A)] applies the A-weighting filter to approximate the frequency sensitivity of the human ear and is the standard regulatory metric for vehicle pass-by noise and interior noise targets [13,18]. Overall SPL [dB(A)] comparisons appear throughout this review; typical values cited include reductions of 3–5 dB(A) from porous road surfaces (Section 3.2) and 2–3 dB prediction accuracy in LBM–SEA wind noise models (Section 3.3). Vibration quantities are typically reported as acceleration levels in dB re 1 μm/s² or as velocity levels, with structural damping characterized by the loss factor η (dimensionless) or the equivalent damping ratio ζ.

Psychoacoustic metrics characterize subjective perception beyond what A-weighted SPL alone captures. Loudness (N), measured in sones, quantifies the perceived magnitude of sound intensity using Zwicker’s model (ISO 532-1) and is particularly sensitive to the mid-frequency band (1–4 kHz) most important for automotive NVH [25]. Sharpness (S), measured in acum, reflects the perceived high-frequency content of a sound and increases with the proportion of spectral energy above approximately 1 kHz; high sharpness is a principal perceptual complaint for EV motor whine [10,22]. Roughness (R), measured in asper, quantifies amplitude modulation in the 15–300 Hz modulation frequency range and is the primary metric for characterizing gear rattle, torsional oscillation, and inverter-induced motor noise texture [25]. Fluctuation Strength (F), measured in vacil, captures slow (<20 Hz) amplitude modulations associated with low-frequency idling and accessory noise. Tonality (K), or tonal prominence, assesses the degree to which discrete tonal components stand above a broadband masking noise floor and is critical for EV-specific NVH evaluation where tonal signatures are not masked by combustion broadband noise [11,12,22]. Psychoacoustic Annoyance (PA) combines loudness, sharpness, and roughness into a single perceptual discomfort predictor and is increasingly used as a composite design target in EV sound quality programs [25,26,27].

Speech intelligibility metrics are relevant to autonomous vehicles where cabin acoustic conditions must support reliable voice-command human–machine interfaces. The Articulation Index (AI) and its successor the Speech Intelligibility Index (SII, ANSI S3.5) quantify the proportion of the speech signal that is audible above the noise floor in weighted frequency bands, with values of 1.0 representing perfect intelligibility and values below 0.3 indicating severely degraded communication [30,31]. Transmission Loss (TL), typically reported in dB as a function of frequency, characterizes the sound insulation performance of panels, barriers, and acoustic packages, and is central to the countermeasure selection methods discussed in Section 7.1 and Section 7.2. Together, these metrics span physical acoustic measurements, psychoacoustic descriptors, and intelligibility indices, providing a consistent evaluation framework for comparing noise reduction performance across the diverse technologies reviewed in this paper.

3. Major Noise Sources in Automobiles

3.1. Powertrain Noise

Powertrain NVH can be separated into combustion-related excitation, mechanical friction sources, accessory drive noise, and transmission/driveline contributions. Early internal combustion engine (ICE) studies focused on reciprocating mechanics and combustion noise radiation [32,33,34,35]. Gear whine—one of the most prevalent tonal NVH issues—has been modeled extensively via transmission error (TE) theory, tooth stiffness variation, and gear mesh dynamics [36,37,38,39,40].

Modern works investigate lighter-weight and downsized turbocharged engines, which introduce challenges such as increased structural sensitivity and high-frequency tonal content [41,42,43,44]. Transmission-related NVH studies highlight torsional vibration, gear rattle under low-load conditions, clutch engagement harshness, and driveline shudder [45,46,47,48]. With hybrid powertrains, additional noise arises from electric motor inverters, harmonic torque ripple, and planetary e-drive gear sets [49,50,51]. The EV-specific implications of these phenomena are discussed at greater depth in Section 8, where electromagnetic excitation mechanisms and their interaction with structural modes are examined.

Table 1. Key powertrain noise sources and primary excitation mechanisms.

Noise Source	Primary Mechanism	Key References
Combustion noise	Pressure oscillation in cylinder, block radiation	[32,33,34,35]
Gear whine	Transmission error, tooth stiffness variation	[36,37,38,39,40]
Gear rattle	Backlash-driven impacting under low load	[45,46]
Driveline shudder	Torsional oscillation, clutch friction dynamics	[47,48]
Inverter/motor noise (HEV/EV)	PWM switching harmonics, torque ripple	[49,50,51]
Motor/inverter housing radiation (HEV/EV)	Structural response of motor/gearbox housing to electromagnetic and bearing forces; surface radiation of flexural waves as airborne noise	Refs. [49,50,51,52,53,54,55]; multi-physical FEM-electromagnetic coupling analysis recommended

summarizes the key powertrain noise sources and their primary generation mechanisms.

3.2. Tire–Road Noise

Tire–road noise is increasingly recognized as the dominant source of exterior and interior noise above approximately 40–50 km/h. Its generation involves complex interactions between tread block impact, air pumping, sidewall vibration, and cavity resonance [56,57,58,59,60]. The seminal work by Kropp [61] and impact–vibration behavior studies by Hamet and Guyader [62] established much of the modern theoretical foundation. The 200 Hz tire cavity mode, well-documented in automotive studies [63,64], is a persistent NVH concern, especially for EVs, due to the reduction in broadband masking noise from the powertrain.

Tire noise generation is governed by two primary mechanisms: structural vibration of the tire belt and sidewall driven by road surface asperities, and aerodynamic mechanisms including air pumping, horn effect amplification at the contact patch, and resonant air column excitation within tread grooves [56,57,58,59,60]. The relative contribution of each mechanism depends on road texture wavelength, vehicle speed, and tire construction. At speeds above approximately 60 km/h, aerodynamic mechanisms progressively dominate, while below this threshold structural excitation from tread block impact and tire cavity resonance are paramount [61,62,63,64]. The horn effect warrants specific explanation: the geometry of the tire–road contact patch forms a diverging horn-shaped channel whose cross-sectional area increases from zero at the contact line to the open road surface. This geometry amplifies sound radiation in a frequency-dependent manner determined by the horn cut-off frequency f_c ≈ c/(π·a), where c is the speed of sound and a is the half-width of the contact patch (typically 70–90 mm for passenger car tires). Below f_c (approximately 1.2–1.5 kHz), the horn provides geometric gain of up to 10–15 dB relative to free-field radiation; above this frequency the amplification decreases as the acoustic wavelength becomes shorter than the horn dimensions [56,61]. Road macro-texture wavelengths in the 5–20 mm range excite the horn mechanism most strongly, explaining why coarse aggregate surfaces are disproportionately noisier than their roughness alone would predict. Porous asphalt mitigates horn amplification by providing a distributed absorptive termination that disrupts the coherent pressure build-up at the contact edges [65]. These distinctions have important design implications: structural noise requires modification of tire stiffness distribution and damping, while aerodynamic noise demands tread pattern optimization and contact patch geometry management.

The acoustic resonance of the tire cavity—the toroidal air volume enclosed between tire and wheel—typically occurs near 200–230 Hz and manifests as a sharp tonal peak in the vehicle interior noise spectrum [63,64]. This mode couples strongly to the wheel and suspension system, transmitting structure-borne energy into the body-in-white. Conventional mitigation strategies include cavity-mounted Helmholtz resonators integrated into the wheel rim, tire foam fillers, and tuned rubber absorbers; recent studies have demonstrated up to 8 dB attenuation of the cavity mode through optimized resonator placement within the wheel cavity [64]. An emerging and promising alternative is the application of acoustic metamaterial (AMM) liners and coiled-up resonator arrays. Three-dimensionally printed coiled-up space resonators—in which the acoustic path is folded within a sub-wavelength volume by a labyrinthine internal geometry—can achieve broadband low-frequency absorption with liner thicknesses of only 15–25 mm, substantially thinner than the foam or passive resonator assemblies they replace [66,67,68]. This sub-wavelength thickness benefit is particularly valuable in the constrained geometry of the wheel well. At 200–230 Hz, a conventional quarter-wavelength absorber would require approximately 375 mm of depth; the coiled geometry reduces this requirement by a factor of 5–10. Additive manufacturing enables customized AMM geometries tailored to the specific resonance frequency and bandwidth of a given vehicle’s tire cavity mode, with recent demonstrations showing insertion loss of 6–12 dB in the 180–250 Hz band [66,67,68]. In EVs, where conventional powertrain noise is absent and tire cavity resonance becomes perceptually dominant, AMM-based liners represent a mass-competitive and packaging-efficient alternative to conventional foam fillers that warrants accelerated development and production validation.

Road surface texture exerts a dominant influence on tire rolling noise. Macro-texture wavelengths (0.5–50 mm) drive impact and air pumping mechanisms, while micro-texture affects friction and contact stiffness [65,69]. Porous and drainage asphalt surfaces can yield noise reductions of 3–5 dB compared with dense bituminous surfaces, primarily by attenuating the horn-effect amplification at the contact patch and providing additional sound absorption [65]. Standardized measurement procedures such as ISO 11819 (Statistical Pass-By) and the close-proximity CPX method are widely used to characterize road-dependent noise levels [70,71]. Computational modeling of tire noise has advanced considerably with wave-based approaches using periodic finite element models of the tire cross-section, and the TNO SWIFT tire model extending standard contact mechanics to include contact patch dynamics for road noise prediction [72,73,74,75,76,77].

The academic literature reviewed in this section demonstrates consistent progress toward multi-physics tire noise prediction; however, unified validated models that simultaneously address structural, aerodynamic, and cavity mechanisms across the full speed range remain an outstanding research need. Machine learning surrogate models have recently been applied to accelerate tire design optimization, reducing the search space over tread geometry and compound parameters by orders of magnitude [78,79,80,81,82]. These developments are discussed in the context of broader data-driven NVH methods in Section 10.2.

Active road noise cancellation (RNANC) represents a rapidly maturing technology for addressing tire-road noise at the receiver rather than the source. Unlike ANC systems targeting periodic powertrain harmonics, road noise is broadband, non-stationary, and highly speed-dependent, presenting substantial algorithmic challenges. Classical filtered-X Least Mean Squares (Fx-LMS) controllers, while effective for stationary periodic noise, exhibit convergence instability and residual errors under non-stationary road noise conditions, particularly during acceleration, braking, and road surface transitions [83,84,85]. Recent algorithmic advances specifically targeting non-stationary road noise include: (1) Variable step-size Fx-LMS (VSS-Fx-LMS) algorithms that adapt the update step size in real time based on estimated signal non-stationarity, maintaining fast convergence during transients while ensuring stability on stationary surfaces [85,86]; (2) Kalman filter-based reference signal predictors that compensate for the delay between the reference accelerometer (typically on the wheel hub or suspension upright) and the error microphone, a critical challenge at higher vehicle speeds where the acoustic transfer path decorrelates from the structural reference; (3) deep reinforcement learning controllers trained on diverse road surface and speed profiles that learn optimal control policies without explicit filter update rules, demonstrating superior performance on variable-speed profiles compared with fixed-architecture adaptive filters [82,87]; and (4) model predictive RNANC frameworks that incorporate vehicle speed and GPS-based road texture previews from connected vehicle data to pre-adapt filter coefficients before road surface transitions occur, reducing onset transients. Production RNANC systems achieving 3–6 dB broadband attenuation in the 100–500 Hz band are now commercially deployed in premium EVs, and algorithmic improvement specifically for variable-speed driving cycles is an active research focus across both OEM and academic NVH communities [85,86,87].

3.3. Wind Noise

As powertrains become quieter, aerodynamic noise becomes a key comfort attribute. Wind noise originates from boundary layer turbulence, A-pillar vortices, mirror-induced flow structures, and leak paths through doors and glazing seals [88,89,90,91]. Early aeroacoustic theories from Lighthill [92] and Curle [93] provided the mathematical basis, while later automotive-specific studies expanded understanding of side mirror vortex shedding and seal leakage mechanisms [94,95,96,97]. Computational fluid dynamics (CFD) methods such as Detached Eddy Simulation (DES) and Lattice Boltzmann Methods (LBM) are now common for predicting wind noise excitation and cabin pressure fluctuations [98,99,100,101].

The aeroacoustic environment around a moving vehicle is characterized by a turbulent boundary layer that develops over the vehicle body, separated flow regions around the A-pillars, roof edges, and door mirrors, and vortex shedding from bluff body components [88,89,90,91]. For highway speeds above 120 km/h, wind noise is often the dominant noise source in modern EVs, surpassing both road noise and residual drive-unit noise [6,7,8,9]. The A-pillar vortex, particularly prominent at yaw angles of 5–15 degrees corresponding to real-world crosswind conditions, generates intense low-frequency pressure pulsations in the 100–400 Hz band on the side window [88,89]. A critically important but often underemphasised aspect of this phenomenon is the vibro-acoustic coupling between the A-pillar vortex pressure field and the side window glazing. The side window is not a rigid boundary: it behaves as a fluid–structure-coupled panel whose flexural vibration modes are excited by the fluctuating aerodynamic pressure field. The fundamental bending mode of a typical passenger car side window (area approximately 0.25 m², thickness 4–6 mm) falls in the 150–300 Hz range, well within the A-pillar vortex excitation band [88,101]. Below the acoustic coincidence frequency f_c = c²/(1.8·c_L·h) (where c_L is the longitudinal wave speed in glass and h is thickness, giving f_c ≈ 3–5 kHz for standard glass), the glass panel radiates interior noise inefficiently; however, near its structural resonances it becomes a highly efficient sound radiator due to mass-law violations at coincidence frequencies. Cremer’s structural–acoustic coupling theory [14] provides the classical framework for quantifying this transmission, and has been applied to modern laminated glazing in which a viscoelastic interlayer between two glass plies provides both safety and vibration damping. The viscoelastic interlayer increases the structural damping of the composite glazing assembly to loss factors of 0.05–0.15 at ambient temperature (compared with 0.002–0.005 for monolithic glass), reducing resonance amplification and improving interior wind noise by 2–4 dB in the 200–500 Hz range [96,101]. Aerodynamic mitigation strategies (A-pillar leading edge shaping, vortex generators at the windshield base, mirror geometry optimization) remain the primary engineering levers, complemented by structural-acoustic glazing optimization for frequency-targeted cabin noise reduction.

Lattice Boltzmann Method simulations, as implemented in commercial tools such as Exa PowerFLOW, have become the dominant computational approach for vehicle-level wind noise prediction [99,100]. LBM operates on a mesoscopic kinetic equation framework and inherently resolves turbulent pressure fluctuations across a broad frequency range without the acoustic analogy post-processing required by Reynolds-Averaged Navier–Stokes (RANS)-based approaches. Coupled LBM–Statistical Energy Analysis (SEA) workflows have demonstrated prediction accuracy within 2–3 dB of measured interior wind noise levels, enabling objective comparison of exterior body designs before clay models are constructed [101]. Door and window sealing systems represent a critical wind noise pathway; seal compression force, gap geometry, and material shore hardness all influence the onset of aeroacoustic leakage and flutter [95,96].

3.4. Structural and Body Noise

The vehicle body structure contributes significantly to NVH through panel vibrations, boom noise, and structural modal behavior. Low-frequency boom—often associated with body-in-white (BIW) mode shapes between 30–80 Hz—has been discussed extensively in the literature [102,103,104,105]. Body sealing stiffness, structural joints, and local reinforcements significantly affect acoustic transfer paths [106,107,108]. Metal panels radiate airborne noise via coincidence phenomena, as described in classic vibroacoustic theory [109,110,111,112,113]. Modern lightweight materials (aluminum, composites) exhibit different stiffness–damping characteristics, requiring updated modeling approaches [114,115,116,117]. The conventional response to panel noise—adding mass via bituminous damping pads or barrier layers—directly conflicts with lightweighting targets. Locally multi-resonant acoustic metamaterials (LRAMs) integrated within or attached to BIW panels offer a mass-neutral pathway to address this conflict. LRAMs consist of periodic arrays of sub-wavelength resonators (mass-spring units, membrane resonators, or coiled-up cavities) tuned to the problematic panel resonance frequency. At the local resonance frequency, energy is transferred from the host panel to the resonator array and dissipated, creating a stop-band in which flexural wave propagation is strongly attenuated without net mass addition to the structure [66,67,68]. The stop-band bandwidth is typically narrow (±20–30% of the resonance frequency for standard single-resonator designs), but multi-resonant arrays with distributed tuning frequencies can achieve broader-band attenuation of 10–20 dB across the 100–500 Hz BIW boom frequency range at surface mass densities of 1–3 kg/m², compared with 4–8 kg/m² for equivalent bituminous treatments [66,67,68]. Integration of LRAM elements within the hollow extrusion sections of aluminium or CFRP BIW structures—so the resonators occupy otherwise unused structural voids—represents a particularly promising packaging strategy for achieving mass-law-exceeding panel transmission loss while maintaining overall vehicle mass targets. The NVH implications of lightweight construction are treated in greater depth in Section 7.2 and Section 10.2.

4. Vibration Phenomena and Harshness

4.1. Driveline Vibration

Driveline vibration arises from imbalance, misalignment, torsional oscillations, U-joint nonlinearity, and clutch/gear engagement transients. Foundational works on torsional dynamics and multi-body coupling mechanics include studies by Rabeih & Crolla [118], Nestorides [119], and subsequent analytical torque ripple models [120,121]. Shudder and surge events are extensively analyzed in automatic and dual-clutch transmissions, where hydraulic control, clutch friction characteristics, and driveline compliance interact nonlinearly [122,123,124,125].

4.2. Idle Vibration and Body Boom

Engine idle shake is governed by low-frequency excitation transmitted through mounts and subframes. Optimization of mount stiffness/damping and powertrain rigid-body modes has been historically central to NVH development [126,127,128,129,130,131,132]. Dynamic characterization of elastomeric engine mounts—including geometry-dependent stiffness and damping behavior—provides essential input data for these optimization studies [132]. Ride quality evaluation across passive and semi-active suspension configurations has further demonstrated that suspension system design choices have a direct and measurable influence on interior NVH comfort [130,131,133]. Body boom issues occur when engine- or road-induced excitations coincide with cabin cavity resonances or BIW flexible modes [134,135,136]. Multi-body dynamic (MBD) simulation techniques described in Section 6.4 are routinely applied to characterize these coupled system resonances and evaluate mount configuration alternatives in a virtual development context.

4.3. Transient Harshness

Harshness refers to short-duration, high-rate-load events such as pothole impacts, brake judder, shift shocks, and suspension bottoming [137,138,139,140]. Brake squeal—a tonal friction-induced instability originating at the pad–disc interface—represents a closely related but distinct NVH concern, characterized using response surface and finite element methodologies [141]. Experimental investigation methods rely on road simulators, high-speed data acquisition, and multi-body simulation correlation [142,143,144]. EVs are particularly susceptible to transient harshness because of their quiet baseline noise floor, which removes broadband masking that would otherwise render transient events less perceptible [143,144]. A dimension of transient harshness that deserves specific attention in the context of lightweight vehicle architectures is the excitation of high-order body-in-white (BIW) structural modes. A conventional steel BIW at vehicle level exhibits high modal density above approximately 80–100 Hz: many closely spaced modes overlap in frequency, their responses averaging out and producing a relatively smooth, low-amplitude transient harshness signature. In a lightweight architecture using aluminium extrusions, CFRP panels, or multi-material sandwich structures, the reduced mass raises the fundamental BIW modes to higher frequencies while simultaneously reducing the modal overlap factor (number of modes per frequency bandwidth). The result is sparser, more isolated mode peaks that respond more sharply to impulsive inputs, producing higher peak accelerations at occupant contact points (seat rail, steering wheel, floor) during pothole traversal or road impact events [143,144]. This mechanism means that lightweight vehicles with equivalent static stiffness to their steel counterparts can exhibit objectively worse transient harshness, a counter-intuitive penalty that is not captured by quasi-static stiffness targets and requires modal density analysis over the full vehicle operational frequency range. Multi-body dynamic simulation coupled with flexible BIW FEM models is essential for predicting this behavior before physical prototype stages. This EV-specific harshness sensitivity is addressed in the broader context of EV NVH challenges in Section 8.

Collectively, the vibration phenomena reviewed in this section underscore the importance of coupled powertrain–chassis–body system modeling. The transition from single-discipline analysis to fully integrated multi-physics simulation—described in Section 6—has been the primary enabler of improved harshness prediction and countermeasure development over the past two decades.

5. NVH Measurement and Testing Techniques

5.1. Modal Testing and Operational Modal Analysis

Modal testing using hammer excitation, shakers, or operational inputs is foundational for NVH diagnostics. The key methodologies are described by Ewins [16], He & Fu [145], and Richardson [146]. Operational Modal Analysis (OMA) enables mode extraction from road-load excitation without controlled input forces and has gained wide acceptance in full-vehicle NVH development due to its practical advantages over classical impact or shaker excitation [147,148].

Semi-anechoic chambers with 4-poster or 7-poster road simulators represent the core NVH testing environment at most automotive OEMs. These facilities allow controlled reproduction of road surface profiles in a quiet, weather-independent setting while measuring acoustic and vibration responses across the full frequency range of interest (typically 20 Hz–2 kHz) [149,150]. Modern 4-poster rigs incorporate servo-hydraulic actuators with high-force, high-bandwidth capability to faithfully replicate road surface profiles measured on reference test tracks. Correlation between simulator results and on-road measurements requires careful road input characterization, accounting for tire enveloping and nonlinear suspension behavior. Virtual spindle force targets derived from full-vehicle MBD simulations are increasingly used as rig drive signals, enabling hardware-in-the-loop integration between physical test and simulation [151,152,153].

Wind tunnel testing remains essential for aeroacoustic development. Automotive acoustic wind tunnels feature very low background noise (typically below 65 dB(A) at 140 km/h), moving ground planes, and integrated microphone array systems for source localization [154,155]. Advanced wind tunnels now incorporate phased microphone arrays of 150–500 elements enabling real-time beamforming source maps, allowing engineers to identify and rank wind noise contributors on the vehicle surface at each design iteration [154,156]. Integration of wind tunnel testing with LBM–SEA computational predictions in a closed-loop development process enables faster convergence to program noise targets while reducing physical tunnel time. An important and often underappreciated limitation in wind tunnel cabin assessment is aeroacoustic leakage: at high speeds, even well-sealed test vehicles exhibit micro-leakage pathways through HVAC ducting, door seal gaps, and cable penetrations that allow external aeroacoustic pressure fluctuations to enter the cabin, artificially elevating the interior noise floor and masking the true structural transmission performance of the body. In production development, aeroacoustic leakage at highway speeds can contribute 2–5 dB(A) to the cabin noise budget in the 500–2000 Hz range, degrading the signal-to-noise ratio for subjective panel assessments and obscuring the effectiveness of structural NVH countermeasures under evaluation [95,96]. Conventional mitigation relies on seal geometry optimization and positive-pressure cabin pressurization; however, emerging research applies AMM-based ventilated duct designs—in which labyrinthine acoustic metamaterial inserts line HVAC ducts and cable penetration bushings—to achieve broadband transmission loss of 10–20 dB without restricting airflow, directly addressing leakage-driven noise at its structural ingress points [66,67,68]. These AMM duct solutions enable cabin assessments with reduced leakage contamination, improving the measurement validity of structural countermeasure evaluations at the test facility.

5.2. Acoustic Intensity, Beamforming, and Holography

Sound intensity measurements allow identification of noise source strength and energy flow paths [157,158,159]. Beamforming and acoustic holography provide high-resolution visualization of noise sources under wind-tunnel conditions or on-road environments [154,155,156,160]. Modern spherical beamforming arrays enable 3D localization with improved spatial resolution compared to planar array configurations [161,162,163]. These capabilities are increasingly employed alongside transfer path analysis to provide an integrated diagnosis of both source contributions and transmission pathway characteristics.

5.3. Transfer Path Analysis (TPA)

TPA has become an essential tool for separating structure-borne and airborne contributions to cabin noise. Classical TPA, operational TPA, and mount-frequency response (FRF) TPA methods are well covered in the literature by Verheij [164], de Klerk & Ossipov [165], and Janssens et al. [166]. The method is widely used to optimize mount configurations, isolate problematic excitations, and modify transfer paths in both physical tests and virtual environments. Neural network-based TPA estimation approaches have recently been proposed to accelerate path contribution analysis from large multi-channel measurement datasets [82]. A significant methodological advance that has become the current industry standard for e-drive NVH development is Component-based TPA (also termed Virtual TPA), in which active sources are characterized independently of the receiver vehicle using blocked forces or equivalent source descriptors measured on a test bench, rather than in the assembled vehicle [164,165,166]. This source characterization is receiver-independent: the blocked forces of a given e-drive unit, measured once on a compliant test rig, can be combined with the body structure’s transfer functions (measured or simulated) to predict the contribution of that e-drive to the interior noise of any receiver vehicle without re-testing the source in each new installation. This approach is particularly powerful for electric drive NVH development where the e-axle is a supplier-developed module integrated into multiple vehicle platforms: source forces are measured at high rotational speeds (up to 20,000 RPM) in a controlled bench environment, avoiding the masking from wind noise and road inputs that confound in-vehicle characterization, and the resulting source descriptor is portable across vehicle programs [82,164,165,166]. Virtual TPA further extends this to fully simulation-based transfer functions, enabling e-drive contribution prediction at vehicle level before any physical prototype exists.

5.4. Buzz, Squeak, and Rattle (BSR) Testing

BSR is a major quality concern driven by lightweight interiors and reduced masking noise. Objective BSR detection methods include structure-borne accelerometer monitoring, digital stethoscopes, and acoustic signature classification using machine learning [167,168,169,170]. Environmental chambers, road simulators, and dedicated BSR rigs are now standard in OEM NVH laboratories [171,172,173].

BSR noise is defined as any unwanted interior sound arising from contact, friction, or impact between adjacent trim components under dynamic loading. As vehicle cabins become quieter and interior trim assemblies more complex, BSR has become one of the leading customer quality complaints, particularly in premium and EV segments. The challenge is compounded by strong sensitivity of BSR onset to material aging, temperature variation, humidity, and manufacturing tolerances [167,168,169,170]. Machine learning approaches for automated BSR detection have advanced rapidly; convolutional neural networks trained on labeled acoustic and accelerometer data have demonstrated classification accuracy exceeding 90% for distinguishing BSR event types (squeak vs. rattle vs. tick) in controlled rig environments [79,169]. Deep learning models trained on fleet data have further been applied to predict in-field BSR occurrence probabilities from vehicle usage profiles, enabling proactive quality interventions before customer delivery [81].

The measurement and testing techniques reviewed in this section collectively provide the experimental foundation for NVH model validation, design decision-making, and production quality assurance. The increasing integration of these techniques with computational models—described in Section 6—and with machine learning diagnostics—addressed in Section 10—represents the defining trajectory of modern NVH development practice.

Table 2. Summary of NVH testing methodologies and primary applications.

Method	Primary Application	Key References
Modal testing/OMA	Mode shape extraction, model correlation	[16,145,146,147,148]
4-/7-poster road simulator	Road noise, harshness, durability	[149,150]
Acoustic wind tunnel	Wind noise source ranking	[154,155]
TPA	Path contribution analysis, mount optimization	[164,165,166]
Beamforming/Holography	Source localization and visualization	[154,155,156,160,161,162,163]
BSR rig/environmental chamber	Squeak, rattle, buzz detection	[167,168,169,170,171,172,173]

provides a summary of the principal testing methodologies and their primary application domains.

6. Computational Modeling Approaches

6.1. Finite Element Method (FEM)

FEM is the backbone of structural NVH prediction. Automotive applications include BIW modal prediction, panel vibration, mount stiffness tuning, and full-body acoustic–structural coupling [174,175,176,177,178,179]. Modern NVH workflows integrate detailed FE models with optimization and sensitivity analysis to identify structural reinforcements and weight reductions [178,179,180,181].

Modern NVH FEM workflows have evolved substantially beyond traditional modal and harmonic analysis. Reduced-order modeling (ROM) techniques, including Craig-Bampton and Rubin substructuring, allow large complex FE models to be condensed into compact superelements that can be assembled and solved in seconds rather than hours, enabling Monte Carlo analyses for variability assessment and rapid design space exploration [177,178,179,180]. Component mode synthesis (CMS) permits individual subsystems—engine, gearbox, suspension, body—to be developed independently and assembled at vehicle level without re-running full body models. The computational cost of full-vehicle NVH FEM, once requiring overnight compute runs, is now routinely completed in under two hours on modern HPC clusters, enabling same-day design iterations [105,179]. Iterative optimization algorithms embedded within commercial NVH preprocessors can automatically identify panel thickness, spot weld density, and bead pattern modifications that minimize target point responses while satisfying mass constraints.

6.2. Boundary Element Method (BEM) and Hybrid FEM/BEM

BEM is preferred for exterior acoustic radiation because it requires only surface discretization, reducing problem dimensionality compared with domain-based methods. Hybrid FEM/BEM coupling is widely used for predicting interior cabin acoustics and panel radiation [182,183,184,185,186]. The literature highlights the importance of solving high-frequency problems using fast multipole BEM and statistical methods at higher frequency bands [187,188,189,190].

6.3. Statistical Energy Analysis (SEA)

SEA is essential in mid-to-high-frequency NVH prediction. Lyon’s foundational work [15] formed the basis, with automotive extensions developed through subsystem characterization and experimental validation [191,192,193,194,195]. SEA models are especially valuable for vehicle interior road noise and wind noise predictions, where the density of structural modes precludes feasible deterministic FEM analysis.

SEA has been extended beyond its classical mid-frequency application domain through hybrid deterministic–statistical methods that bridge the frequency gap between FEM (accurate below approximately 500 Hz) and pure SEA (valid above approximately 1–2 kHz). The hybrid FEM–SEA method, implemented in tools such as VA One (ESI Group), models low-modal-density components deterministically while treating high-modal-density panels statistically, covering the mid-frequency range (500–1500 Hz) that is acoustically critical for cabin comfort but numerically challenging for either method alone [191,192,193,194,195]. Uncertainty quantification in SEA models, acknowledging manufacturing variability in panel properties and connections, has received growing research attention, with ensemble SEA approaches quantifying prediction confidence bounds [192,193].

6.4. Multi-Body Dynamics (MBD)

For driveline, suspension, and powertrain NVH, MBD simulations provide dynamic behavior analysis under transient loads. Studies incorporate nonlinear bushings, joint clearances, flexible bodies, and tire models [76,77,196,197,198,199]. Coupled MBD–FEM modeling is now commonplace for transient road harshness and shift event simulations [151,152,153], enabling prediction of harshness responses at occupant interfaces without reliance on physical prototypes at intermediate development gates.

6.5. Computational Aeroacoustics (CAA)

Wind noise prediction has progressed through DES, LBM, and hybrid RANS–CAA workflows [200,201,202,203]. CAA simulations provide detailed pressure fluctuation maps required for cabin-side acoustic modeling [204,205,206]. Physics-informed machine learning (PIML) represents an emerging computational paradigm that incorporates governing equations of structural acoustics as constraints within neural network training, improving generalization beyond the training data distribution [78,79,80,81,82]. Surrogate modeling using Gaussian process regression, neural networks, and polynomial chaos expansion is now routinely applied to construct fast-running NVH response emulators from high-fidelity FEM simulation databases. The convergence of CAE simulation, experimental measurement, and machine learning within integrated digital workflows is a defining trend in contemporary automotive NVH development, discussed further in Section 10 and Section 11.

The computational methods reviewed in this section collectively cover the frequency spectrum from low-frequency driveline and structural resonances (MBD, FEM) through mid-frequency cabin acoustics (hybrid FEM–SEA) to high-frequency panel radiation and aeroacoustics (BEM, LBM, SEA). Sustained integration of these methods within multi-fidelity simulation workflows—where model fidelity is matched to the frequency domain of interest—is the characteristic feature of world-class virtual NVH development programs [151,152,153,174,175,176,177,178,179,180,181,182,183,184,185,186,187,188,189,190,191,192,193,194,195,196,197,198,199,200,201,202,203,204,205,206].

Table 3. Comparative analysis of computational NVH modeling methods: frequency applicability, advantages, limitations, and primary use cases.

Method	Frequency Range	Key Advantages	Limitations	Primary NVH Applications
FEM	<500 Hz	High spatial resolution; handles complex geometry; deterministic; supports topology optimization	Mesh density escalates rapidly with frequency; computationally prohibitive above ~500 Hz at vehicle scale; poor at representing statistical variability	BIW modal prediction; mount stiffness tuning; low-frequency structural-acoustic coupling; panel reinforcement optimization
BEM/Hybrid FEM-BEM	100 Hz–2 kHz	Surface-only discretization; suited for exterior radiation and interior cavity acoustics; naturally satisfies radiation condition	Dense system matrices; standard BEM breaks down near structural resonances; computationally expensive at high frequency without fast multipole acceleration	Panel radiation prediction; interior cabin acoustics; NVH target setting via acoustic sensitivity maps
SEA/Hybrid FEM-SEA	500 Hz–10 kHz	Computationally efficient at high frequencies; handles statistical variability well; naturally suited to diffuse field environments	Requires high modal overlap (typically >3 modes per band); poor accuracy for low-modal-density subsystems; subsystem coupling loss factor estimation is a persistent challenge; not suitable for transient or tonal predictions	Road noise and wind noise interior predictions; acoustic package design; mid-to-high frequency body structure
MBD	0–200 Hz (transient)	Handles nonlinear joints, clearances, and friction; time-domain transient simulation; naturally captures gear rattle and clutch dynamics	Limited to rigid or simplified flexible bodies unless coupled to FEM; bushing and joint models require extensive experimental characterization; step-size sensitivity in nonlinear time integration	Driveline torsional dynamics; suspension harshness; powertrain mount optimization; gear rattle and clutch shudder prediction
LBM/CAA	100 Hz–5 kHz	Resolves turbulent pressure fluctuations directly; no acoustic analogy post-processing; predicts broadband and tonal wind noise simultaneously; massively parallelizable on GPU clusters	Very high computational cost (full-vehicle LBM runs require HPC); requires smooth surface CAD data; limited to steady-state or quasi-steady aeroacoustic scenarios; grid-dependent accuracy at low Mach number	A-pillar and mirror wind noise; door seal leakage; cabin pressure fluctuation; exterior aeroacoustic source ranking
ML Surrogates/PINNs	Broadband (data-dependent)	Orders of magnitude faster than physics solvers once trained; enables real-time optimization and Monte Carlo uncertainty quantification; PINNs embed physical constraints to reduce training data requirements	Accuracy limited to training distribution; large labeled datasets required for standard ML; extrapolation unreliable; interpretability limited; not yet suitable as stand-alone design tools without physics model validation	BSR classification; tire design optimization surrogates; sound quality prediction; real-time in-vehicle NVH monitoring; fleet data analytics

, presented below, provides a comparative perspective on these methods, synthesizing their respective frequency domains, computational costs, accuracy limitations, and primary application areas to guide method selection in practical NVH development programs.

7. NVH Materials and Countermeasures

7.1. Passive Materials: Damping, Absorption, and Insulation

Traditional NVH control techniques rely on damping treatments, barrier materials, and acoustic absorbers. Viscoelastic damping models are documented by Rao [207] and Nashif [208], while multilayer acoustic barrier principles are described by Allard & Atalla [209]. Porous absorbers, glass fiber mats, foams, and acoustic packages are widely employed for interior noise reduction [210,211,212,213]. Studies show that constrained layer damping (CLD) and tuned mass dampers (TMDs) offer effective solutions for panel and structural resonances [214,215,216,217].

Constrained layer damping (CLD) treatments consist of a viscoelastic polymer layer bonded between the structural panel and a stiff constraining layer. Under bending deformation, shear strain in the viscoelastic layer dissipates energy as heat, providing significant damping augmentation to lightly damped metal panels [214,215]. The damping effectiveness is strongly frequency and temperature dependent, governed by the complex shear modulus of the viscoelastic material. Modern OEM NVH programs use constrained optimization to minimize CLD treatment mass while achieving structural damping loss factor targets at defined operating temperatures. Acoustic package optimization has become a sophisticated multi-domain activity integrating SEA vehicle models with material characterization databases to minimize total acoustic package mass while meeting interior noise targets. Modern acoustic packages employ layered constructions combining barrier layers, decoupler layers, and absorption layers to achieve transmission loss and absorption across the 200–4000 Hz range [209,210,211,212], with demonstrated mass savings of 1–3 kg per vehicle compared with conventional bituminous pad treatments.

7.2. Advanced Materials

Emerging NVH materials include acoustic metamaterials, lightweight acoustic foams, micro-perforated panels, and active–passive hybrid structures [66,67,68,218]. Locally resonant metamaterials consisting of arrays of small resonators attached to or embedded within a host panel can create stop-bands that strongly attenuate flexural wave propagation in targeted narrow frequency bands, addressing discrete tonal NVH concerns such as motor whine or gear mesh harmonics without the mass penalty of broadband treatments [66,67]. The tunability and manufacturing constraints of acoustic metamaterials merit rigorous scientific discussion. The stop-band centre frequency is determined by the local resonance frequency of the individual resonator unit cell, f_r = (1/2π)√(k/m), where k is the local spring stiffness and m is the resonator mass. This enables deterministic frequency targeting, but the stop-band bandwidth is inherently narrow: for a single-resonance design with loss factor η = 0.01–0.05, the −13 dB attenuation bandwidth is typically ±5–15% of f_r. Multi-resonant arrays with distributed unit cell parameters broaden this to ±30–50%, but require precise manufacturing of hundreds of resonator elements per panel to maintain the designed frequency distribution [66,67,68]. Manufacturing constraints are therefore non-trivial: conventional stamping and injection moulding processes cannot economically produce the geometric complexity required for coiled-up resonators at the unit cell scale (typically 20–50 mm), and production implementation currently relies on additive manufacturing (selective laser sintering, fused deposition modelling) which introduces cost and throughput challenges at automotive volume. The adoption pathway to mass production is an open research area, with composite overmoulding and multi-shot injection moulding identified as candidate scalable processes [66,67,68]. Micro-perforated panels (MPPs) offer a complementary lightweight absorber technology: a rigid panel with sub-millimetre diameter perforations acts as a viscous absorber without requiring porous infill, enabling acoustic integration into visible trim surfaces with broadband absorption coefficients of 0.5–0.9 in the 500–4000 Hz range [218,219]. Application of additive manufacturing to produce complex metamaterial geometries and graded-porosity acoustic absorbers is emerging as a route to customised NVH solutions at competitive cost [66,67,68]. Beyond acoustic performance, the relationship between NVH exposure levels and human physiological and psychological response is an important context for evaluating the benefits of advanced materials. Sustained exposure to interior noise levels above 65 dB(A) is associated with increased stress biomarkers, concentration impairment, and driver fatigue, while transient harshness events above 80 dB(A) peak contribute to startle responses [220]. Advanced materials that reduce cabin noise levels by 3–8 dB(A)—as demonstrated by optimised LRAM panels and AMM liners in laboratory conditions—thus offer health and safety co-benefits alongside comfort improvements, a dimension that strengthens the justification for their inclusion in lightweight vehicle architectures [66,67,220].

Composite and multi-material body structures present unique NVH challenges compared to conventional steel bodies that arise directly from the fundamental stiffness–weight–damping trade-off inherent to advanced engineering materials. Carbon fiber reinforced polymer (CFRP) panels are an instructive case: their outstanding specific stiffness (stiffness-to-density ratio approximately five times that of steel) and low areal mass make them highly attractive for lightweighting, but these same properties produce severely adverse NVH behavior through two coupled physical mechanisms. First, CFRP exhibits inherently low structural damping: the loss factor η of unmodified CFRP typically ranges from 0.001 to 0.003, compared with 0.01 to 0.02 for steel panels treated with constrained damping layers, and up to 0.1 for viscoelastic foam composites [208,214]. The damping mechanism in metals arises from thermoelastic coupling and grain boundary micro-plasticity—mechanisms absent in carbon fiber–epoxy composites. Consequently, once vibrational energy is introduced into a CFRP panel, it persists far longer and is radiated as sound far more efficiently than in equivalent steel panels. Second, the high specific stiffness of CFRP raises the panel’s natural frequencies, pushing resonances into the mid-frequency range (500–2000 Hz) that is most perceptually sensitive and most difficult to address with passive treatments. By contrast, a thicker, heavier steel panel with the same overall stiffness will have lower resonant frequencies and higher radiation damping owing to its greater mass. The net result of substituting steel with CFRP at equivalent stiffness is thus a lighter panel that vibrates more at higher, more audible frequencies, for longer, and radiates more sound per unit of input excitation—a triple NVH penalty despite the mass reduction. Hybrid material joints between aluminum extrusions, CFRP panels, and steel reinforcements further introduce local stiffness discontinuities that act as vibrational impedance mismatches, reflecting wave energy and complicating transfer path analysis [221,222]. Bonded adhesive joints provide significant structural damping at bond interfaces—loss factors of 0.05–0.15 have been reported for stiff epoxy adhesives under shear loading—partially compensating for reduced base material damping, but only within narrow temperature and frequency ranges that require careful characterization [221,222]. The NVH implications of lightweighting continue to be an active research area; key challenges and opportunities are identified in Section 10.3 as a priority for future investigation.

8. NVH in Electric Vehicles

Electric vehicles introduce new NVH challenges due to near-silent operation at low speeds. The absence of engine masking amplifies perception of secondary noise sources such as tire cavity resonance, inverter switching harmonics, and gearbox whine [223,224,225,226]. Permanent magnet synchronous motors (PMSMs) generate electromagnetic NVH through radial forces, cogging torque, and stator slot harmonics [52,53,54,55]. EV-specific literature also highlights the emergence of “whine bands” and tonal phenomena linked to PWM frequencies and motor–driveline coupling [227,228,229,230]. Further concerns include battery pack rattle, cooling system noise, and ancillary pump vibrations [5].

The NVH character of battery electric vehicles differs fundamentally from internal combustion counterparts not only in the absence of combustion excitation but also in the frequency content and psychoacoustic character of the dominant noise sources. Electric drive units produce high-frequency tonal noise at harmonics of the electrical drive frequency and motor pole-pair order, typically in the range 1–8 kHz [223,224,225,226]. These tonal signatures, while objectively lower in overall sound pressure level than ICE noise, are perceived as intrusive by occupants accustomed to broadband masking. Psychoacoustic annoyance modeling specifically calibrated for EV tonal environments is therefore a priority research area, with tonality and roughness metrics proving more predictive of subjective ratings than A-weighted levels [5,10,11,12,22,23,24,25,26,27].

Electromagnetic noise generation in permanent magnet synchronous motors (PMSMs) arises from the interaction between stator magnetomotive force (MMF) harmonics and rotor permanent magnets. The dominant electromagnetic excitation orders are determined by the number of stator slots and rotor pole pairs; mismatches between force spatial orders and structural mode shapes determine whether resonances occur within the operating speed range [52,53,54,55]. Motor NVH optimization therefore involves co-design of electromagnetic and structural characteristics—an interdisciplinary challenge spanning electrical engineering and structural mechanics [52,53,54,55,229]. Skewed rotors and distributed winding configurations can reduce spatial harmonics but introduce manufacturing complexity and efficiency penalties.

Power electronics and inverter noise represents an increasingly important NVH concern. Pulse-width modulation (PWM) switching at frequencies typically between 8–20 kHz generates switching harmonics that excite motor windings and radiate high-frequency airborne noise [225,227]. Random carrier frequency modulation (RCFM) and spread-spectrum switching strategies spread PWM energy over a wider bandwidth, reducing tonal prominence [227,228]. E-axle NVH development employs transmission error-based analysis while additionally accounting for electromagnetic stiffness contributions and the high operational speed range—up to 20,000 RPM in some designs—that extends gear mesh harmonics well into the audible tonal range [226,229,230]. Thermal management of e-axle units also influences NVH through temperature-dependent bearing preload and gear backlash changes.

Acoustic Vehicle Alerting Systems (AVAS) are now mandated in many jurisdictions for EVs operating below approximately 20 km/h [231,232]. AVAS sound design is governed by UN Regulation 138, specifying minimum sound pressure level requirements across the 160–4000 Hz band. Beyond regulatory compliance, OEMs are investing in AVAS as a brand differentiation opportunity, and Active Sound Design (ASD) systems extend this concept to higher speeds for driver feedback and sportiness enhancement [231,232,233,234]. The psychoacoustic design of ASD content—balancing authenticity, pleasantness, and informational richness—is an active research area drawing on music cognition, signal processing, and automotive brand studies.

Beyond the drive unit and inverter, three additional high-voltage power electronic subsystems generate significant NVH in contemporary EVs and deserve specific attention. The on-board charger (OBC), which converts AC grid power to DC for battery charging, operates its switching converters at frequencies typically between 50 kHz and 150 kHz; however, inter-modulation between switching harmonics and structural resonances of the OBC housing and adjacent body panels can produce audible tonal noise in the 2–8 kHz range that is particularly perceptible in the quiet EV cabin during stationary charging. The high-voltage DC-DC converter, which steps down the main battery voltage (typically 400 V or 800 V) to the 12 V low-voltage bus, likewise generates switching-frequency electromagnetic interference and associated mechanical vibration of its magnetic components (inductors, transformers) in the 5–20 kHz range [227,228]. Mitigation strategies for both OBC and DC-DC converter NVH include acoustic enclosures, vibration-isolating mounting brackets, spread-spectrum modulation strategies to reduce tonal prominence, and optimized inductor winding geometries to minimize magnetostrictive excitation. The heat pump system, increasingly adopted as the primary cabin thermal management solution in cold-climate EV operation, introduces a distinct NVH signature combining compressor pulsations (fundamental frequency 20–60 Hz and harmonics), refrigerant flow noise (broadband 200–2000 Hz), and expansion valve hissing and clicking transients. Unlike ICE-era HVAC systems where compressor noise was masked by engine sound, the EV heat pump operates in a near-silent acoustic environment, making its NVH behavior a first-order customer quality concern. Active vibration isolation of the compressor mounting and careful refrigerant circuit design to avoid two-phase flow instabilities are among the principal engineering countermeasures.

The EV NVH challenges reviewed in this section demonstrate that electrification has not reduced the overall complexity of vehicle acoustic engineering; rather, it has shifted the dominant problem from low-frequency combustion noise to high-frequency electromagnetic tonalities, elevated the perceptual importance of tire and wind noise, and introduced fundamentally new acoustic design domains in power electronics, battery systems, thermal management, and synthetic sound generation.

Table 4. EV-specific NVH sources, generation mechanisms, and principal mitigation strategies.

NVH Source	Generation Mechanism	Mitigation Strategy
Electromagnetic motor noise	MMF harmonics, radial force waves	Skewed rotors, distributed windings, co-design of EM and structural modes [52,53,54,55]
PWM inverter noise	Switching harmonics 8–20 kHz	RCFM, spread-spectrum switching, acoustic enclosures [227,228]
E-axle gear whine	Transmission error at high RPM	Tighter tooth profile tolerances, active vibration isolation [226,229,230]
Tire cavity resonance	Toroidal cavity mode ~200–230 Hz	Rim Helmholtz resonators, tire foam fillers, tuned absorbers [63,64]
Battery pack rattle	Cell and module contact impacts	Foam cushioning, rigid mounting provisions [5]
Cooling system/pump noise	Fluid-borne and structure-borne excitation	Acoustic enclosures, flexible mounts, anti-vibration hoses [5]

summarizes the principal EV-specific NVH sources, their generation mechanisms, and their associated mitigation strategies.

9. Active Noise and Vibration Control

Active control technologies are increasingly viable as cabin background noise decreases. Active Noise Control (ANC) for low-frequency engine and road noise has been implemented using adaptive algorithms such as the filtered-X Least Mean Squares (Fx-LMS) algorithm [83,84,85,235,236,237,238]. Active vibration control (AVC) and active mounts provide additional attenuation of powertrain and driveline vibration [86,239,240,241]. Intelligent control strategies including LQR and fuzzy logic controllers applied to active suspension systems represent an important class of active vibration management approaches validated on half-car models [242]. EVs benefit significantly from ANC/AVC systems due to dominant low-frequency harmonics and tonal motor noise [5,87,243,244]. Studies on virtual engine sound design also integrate psychoacoustics with active sound generation for improved driving feedback [231,232,233,234].

ANC systems operate by generating anti-noise signals through cabin loudspeakers that destructively interfere with the target noise field at reference microphones near occupant ear positions. The Fx-LMS algorithm, originally formalized by Elliott and Nelson [83], remains the dominant real-time control algorithm due to its computational simplicity and proven convergence properties. Modern implementations utilize multiple-input multiple-output (MIMO) Fx-LMS architectures with four to eight reference microphones and four to twelve error microphones to achieve broadband global attenuation across the cabin [83,84,85]. Feedforward control of periodic noise uses engine order pulses or crankshaft encoders as reference signals, while feedback control architectures address non-periodic broadband noise components.

Road noise ANC is one of the most commercially active application areas, driven by the growing dominance of tire-road noise in both EV and ICE premium vehicles. Broadband road noise ANC systems use acceleration signals from wheel hubs or suspension components as reference inputs to feedforward controllers that adaptively synthesize anti-noise across the 100–500 Hz range [85,86,245]. Attenuation of 3–6 dB is typically achieved in production systems. Challenges include the non-stationary character of road noise with vehicle speed and road surface changes, and the computational burden of MIMO processing in embedded automotive ECUs. Model predictive control and deep reinforcement learning approaches have been proposed to overcome the limitations of fixed-filter Fx-LMS in non-stationary environments [82,87].

Active engine mounts use electrodynamic actuators integrated into rubber isolators to generate counter-forces that cancel mount reaction forces in the 50–250 Hz range, providing attenuation of idle shake and low-speed booming typically exceeding equivalent passive mounts by 10–15 dB in target bands [240,241]. The integration of ANC and AVC with vehicle connectivity and over-the-air (OTA) software update capability opens new possibilities for adaptive NVH management over vehicle lifetime: production ANC systems can continuously retrain their adaptive filters from fleet-collected acoustic data, addressing performance degradation due to speaker aging or cabin configuration variability [87,244].

10. Future Directions and Research Trends

This section synthesizes emerging research directions drawing on both the foundational literature reviewed throughout the paper and, importantly, on recent publications from 2021–2025 that represent the cutting edge of the field. Consistent with best practice for systematic reviews, the reference list for this section has been specifically augmented to ensure adequate representation of recent literature: publications from 2021 onward account for approximately 30% of references in Section 10 and Section 11, reflecting the rapid pace of development in data-driven NVH methods, EV-specific acoustic engineering, and autonomous vehicle cabin acoustics. Where recent primary sources were not yet available for specific sub-topics, the authors have explicitly identified these as open research directions rather than citing older proxy literature.

10.1. Proposed Integrated Research Framework for Next-Generation NVH Development

Based on the systematic analysis of 255 sources presented in this review, and motivated by the research gaps identified throughout, the authors propose an Integrated Multi-Fidelity NVH Development Framework (IMNF) for next-generation vehicle development. This framework synthesizes the principal methodological, technological, and organizational insights emerging from the reviewed literature into a coherent research and engineering roadmap structured around five interconnected pillars.

Pillar 1: Unified Psychoacoustic-Physical Evaluation Metrics. A fundamental gap identified across psychoacoustic (Section 2.2), EV NVH (Section 8), and active control (Section 9) literature is the absence of standardized composite NVH metrics integrating objective physical measurements with calibrated psychoacoustic weighting. Future research should develop and validate multi-dimensional NVH indices combining A-weighted SPL, tonality, roughness, and harshness into deployable target metrics specifically calibrated for EV listening environments, where tonal prominence dominates. Cross-industry standardization efforts analogous to ISO 532 (loudness) are urgently needed.

Pillar 2: Closed-Loop Digital Twin Integration. FEM, SEA, MBD, and ML surrogate models are currently deployed in largely independent silos. The proposed framework advocates for a cloud-connected digital twin architecture in which physics-based NVH models are continuously updated using in-service fleet data. This closed-loop structure enables progressive model refinement, anomaly-triggered design alerts, and lifetime NVH health monitoring—technically feasible based on advances reviewed in Section 10.2 and Section 10.5, but not yet integrated into a production-validated system.

Pillar 3: Electromagnetically Co-Designed Electric Drive Units. EV NVH literature (Section 8) consistently identifies electromagnetic-structural co-design as an unresolved bottleneck. Motor NVH is currently treated sequentially; the proposed framework advocates genuine concurrent co-design in which motor pole-pair count, slot geometry, winding configuration, and housing structural modes are optimized simultaneously within a multi-objective framework including NVH targets alongside electromagnetic performance metrics. This requires new multi-physics optimization tools spanning electrical and mechanical engineering disciplines.

Pillar 4: Adaptive Active NVH Management for Autonomous Platforms. Active control technologies (Section 9 and Section 10.5) are mature at the component level but immature as integrated vehicle-level systems. The framework proposes an onboard Adaptive NVH Management System (ANMS) coordinating ANC, AVC, active mounts, and ASD within a unified real-time model predictive controller with access to predictive road-surface data from autonomous vehicle perception systems, continuously balancing attenuation performance against energy consumption and actuator fatigue.

Pillar 5: NVH-Conscious Multi-Material Structure Design. Lightweighting pressure (Section 10.3) demands NVH engineering with CFRP, aluminum, and multi-material hybrids that have fundamentally different damping and wave propagation characteristics. The proposed framework calls for: (a) validated damping models for bonded joints and hybrid material interfaces; (b) topology and material distribution optimization frameworks jointly minimizing mass and NVH response; and (c) standardized acoustic metamaterial design libraries for targeted tonal attenuation adaptable to new vehicle architectures.

Together, these five pillars constitute a research agenda grounded in the current state of the art documented throughout this review. They address the critical gaps identified across sections and are sufficiently concrete to guide grant applications, doctoral research programs, and industry–academic collaboration initiatives in automotive NVH for the period 2025–2030. The interrelationships among the five pillars—through shared data flows, common modeling tools, and co-optimization opportunities—reflect the fundamentally interdisciplinary character of next-generation NVH development, in which advances in any one pillar create enabling conditions for progress in the others.

10.2. Machine Learning for NVH Prediction

Recent literature highlights the promising role of data-driven methods for source identification, BSR detection, and sound quality evaluation [78,79,80,81,82]. Neural networks and physics-informed machine learning models are being integrated into simulation workflows. Convolutional neural networks (CNNs) applied to spectrogram representations of acoustic measurements enable automated classification of noise source types with accuracy comparable to expert listeners [78,79,80]. Recurrent neural networks and LSTM architectures process time-series vibration data to identify transient events such as gear rattle, bearing defects, and harshness incidents in unstructured driving data [81,82].

Physics-informed neural networks (PINNs) represent a particularly promising approach for NVH surrogate modeling because they embed wave equations and structural dynamic governing equations as soft constraints during network training, ensuring physical consistency even in regions poorly represented by training data [80,82]. Graph neural networks applied to finite element mesh graphs have demonstrated the ability to predict nodal vibration responses without conventional solver runs, enabling interactive design exploration [82]. Digital twin frameworks integrating real-time sensor data from production vehicles with physics-based NVH simulation models enable continuous fleet NVH monitoring and data-driven warranty cost prediction [81,246].

The vehicle NVH digital twin concept deserves more detailed elaboration given its transformative potential for lifecycle NVH management. A fully realized NVH digital twin consists of three coupled components: (1) a validated multi-fidelity physics model of the vehicle’s structural-acoustic system (incorporating FEM subsystem models for key components, SEA representations of high-frequency panels, and MBD models of driveline and suspension); (2) a fleet data ingestion layer that continuously collects accelerometer, microphone, and CAN bus signals from connected production vehicles; and (3) a machine learning inference engine that maps measured sensor patterns to physical model parameter updates, enabling the digital twin to remain synchronized with the actual condition of each individual vehicle throughout its operational life [81,246]. This architecture enables NVH lifecycle degradation prediction through a specific and practically valuable scenario: consider rubber bushing aging in a front suspension subframe. New rubber bushings typically exhibit dynamic stiffness of 500–1500 N/mm in the longitudinal direction and loss factors of 0.10–0.25; after 100,000 km of operation under thermal cycling and load fatigue, dynamic stiffness increases by 20–40% and loss factor decreases by 15–30% as the elastomer crosslink density increases and micro-cracks form [196,197]. In a conventional development process, this degradation is addressed only after customer complaints trigger warranty claims. In a digital twin framework, the NVH model continuously updates its bushing stiffness parameters based on statistical patterns in fleet suspension acceleration data, detects when an individual vehicle’s bushing parameters have drifted beyond a defined NVH-relevant threshold, and triggers a predictive maintenance notification before the occupant perceives any change in road isolation. A parallel scenario applies to battery pack NVH: battery module foam cushioning degrades with thermal cycling, and cell-to-module contact conditions change with charge-discharge-induced expansion and contraction of cells. Digital twin models tracking the acoustic signature of battery pack rattle events—identifiable as characteristic impulsive transients in rear floor accelerometer data correlated with pothole events—can predict foam cushion replacement needs years ahead of structural failure [81]. The convergence of connected vehicle data, machine learning analytics, and physics-based simulation defines the frontier of NVH development capability for the coming decade, with early production implementations already demonstrated in premium EV platforms [81,246].

10.3. Lightweight Vehicles and New Materials

As OEMs pursue aggressive mass reduction strategies, future structures may require new models for damping, joint behavior, and coupling loss [221,222,247,248]. CFRP panels and hybrid metal-composite assemblies introduce low inherent damping and wave reflection sites at material interfaces that conventional SEA and FEM models do not represent with sufficient accuracy. Validated simulation frameworks for multi-material NVH prediction, integrating bonded joint damping models and graded-property composite representations, are identified as a priority research need. Experimental characterization of damping in novel material combinations under realistic dynamic loading and temperature conditions is a prerequisite for model development.

10.4. Automated NVH Tuning and Optimization

Optimization frameworks incorporating genetic algorithms, topology optimization, and design of experiments (DoE) are becoming foundational tools for NVH development [246,249,250,251]. Bayesian optimization frameworks combining Gaussian process surrogate models with efficient acquisition functions automate the search for NVH-optimal design configurations in high-dimensional parameter spaces far more efficiently than conventional DoE screening [246,250]. Generative adversarial networks have been applied to synthesize realistic road noise or wind noise recordings for accelerated subjective evaluation panel studies, providing unlimited acoustic stimulus variation without requiring physical vehicle prototypes [80].

10.5. Autonomous and EV-Dominant Platforms

Autonomous vehicles require extremely low cabin noise for sensor performance and passenger comfort. SAE Level 4 and Level 5 automation concepts envision interior environments comparable to premium quiet spaces, with noise targets potentially below 35 dB(A) at highway speeds in the most demanding scenarios [30,31,252]. Achieving these targets while simultaneously reducing vehicle mass and managing EV-specific noise sources demands a comprehensive rethinking of NVH engineering practice. The acoustic sensing requirements of autonomous driving systems impose additional NVH constraints: ultrasonic parking sensors, LiDAR systems with rotating optical components, and cooling fans for high-performance computing platforms all generate noise that must be isolated from cabin occupants and from speech recognition microphones used for human-machine interaction [30,31,252]. Active noise field management using arrays of cabin microphones and loudspeakers enables spatial control of the acoustic environment in each seating zone independently, creating personalized quiet zones while maintaining audio content in active zones [86,87,244].

The transition to software-defined vehicles (SDVs) and OTA update capability enables NVH management strategies that were previously impossible in fixed-function ECU architectures. AI-powered NVH diagnostics embedded in the vehicle can detect abnormal noise signatures, correlate them with probable mechanical root causes, and trigger predictive maintenance notifications before customer-perceptible degradation occurs [80,81,247]. The integration of NVH management with vehicle energy management systems—trading off active isolation actuator power consumption against noise reduction targets in real time—represents a systems engineering challenge unique to electrified autonomous platforms that will define the research agenda for the next decade.

11. Digital Development and Virtual NVH Engineering

The shift toward virtual-first NVH development has fundamentally changed the organizational and methodological structure of automotive NVH programs. Leading OEMs now target virtual sign-off of NVH performance targets at gate reviews that previously required physical prototypes, relying on validated simulation models to replace or significantly reduce the need for hardware testing at intermediate development stages [151,152,153,174,175,176,177,178,179,180]. This virtual-first philosophy is enabled by continued advances in computational power, mesh quality standards, material characterization databases, and model correlation methodologies. Model correlation and validation remain the critical bottleneck in virtual NVH development. Experimental-analytical correlation (EAC) of full-vehicle NVH FE models requires careful test-analysis consistency in boundary conditions, structural joint modeling, and acoustic source representation. Modal assurance criterion (MAC) analysis quantifies the spatial correlation between measured and predicted mode shapes, with values above 0.9 typically required before a model is considered validated [145,146,147,148].

Virtual acoustic prototyping (VAP) enables auralization—the rendering of acoustic simulation results as audible sound files—allowing engineers and product planners to listen to simulated NVH scenarios before physical hardware exists [10,11,12,22,23,24,25]. Listening panel studies conducted with auralized simulations have demonstrated high correlation with equivalent physical prototype evaluations for comparative assessments, validating VAP as a reliable decision support tool for NVH target setting and concept selection [22,24,25]. The combination of virtual reality head-mounted displays with binaural VAP creates immersive virtual NVH evaluation environments that are becoming standard tools in advanced OEM digital development programs, and represent the leading edge of human factors integration in the NVH engineering process [253].

12. Conclusions

NVH engineering has matured into one of the most complex, technically demanding, and strategically important disciplines in automotive development. This review has documented the breadth of contemporary NVH science, spanning foundational vibroacoustic theory, the characterization of noise sources in ICE, hybrid, and electric powertrains, tire–road and aerodynamic noise mechanisms, advanced measurement and testing methodologies, computational prediction tools from FEM and BEM to SEA and machine learning surrogates, passive and active countermeasure technologies, and the emerging challenges of autonomous and connected vehicle platforms.

Several overarching trends define the current state and near-term trajectory of the field. Electrification has irrevocably shifted the NVH problem from low-frequency combustion and powertrain noise toward high-frequency electromagnetic tonalities, tire cavity resonance, and aerodynamic noise, requiring new psychoacoustic evaluation frameworks calibrated to EV listening environments. Active noise and vibration control has progressed from laboratory concept to production-viable technology, increasingly integrated with premium audio and active sound design systems within a unified cabin acoustic management architecture. Machine learning and digital twin methodologies are transforming every phase of NVH development, from early surrogate-based design screening to fleet-wide health monitoring and predictive maintenance. Advanced materials including acoustic metamaterials, micro-perforated absorbers, and lightweight composite structures are enabling mass-efficient NVH performance levels previously achievable only with heavy passive treatments.

Critical research gaps and opportunities remain. Unified NVH evaluation metrics that integrate objective physical measurements with subjective psychoacoustic perception have yet to be standardized across the industry, limiting consistent target-setting and benchmarking. Addressing this gap requires a multi-layered evaluation framework that bridges three domains: (1) objective multiphysics simulation outputs (SPL spectra, structural vibration FRF, SEA energy levels); (2) standardized psychoacoustic descriptors (Loudness, Sharpness, Roughness, Tonality as defined in Section 2.3); and (3) qualitative subjective response data from structured listening panels, focus groups, and questionnaire instruments designed to capture overall acoustic character, pleasantness, and perceived quality dimensions not reducible to single-attribute metrics [29]. This triangulated approach—analogous to emerging AMM verification methodologies that combine simulation, laboratory acoustic measurement, and human perception testing to validate novel material performance—would enable the NVH community to set targets that are simultaneously physically meaningful, perceptually calibrated, and statistically robust across diverse occupant populations [29]. Real-time in-vehicle NVH monitoring with sufficient spatial and temporal resolution to support closed-loop adaptive control remains an open systems challenge. Validated simulation frameworks for novel EV architectures present a further frontier: axial flux motors, which are gaining commercial traction for their high torque density and flat packaging, generate a fundamentally different NVH signature from the radial flux PMSMs that dominate current e-drive literature. In a radial flux motor, the dominant electromagnetic excitation is a radial magnetic pressure wave acting on the stator bore, exciting the stator ring breathing and ovaling modes in the radial plane. In an axial flux motor, by contrast, the primary magnetic force is axial—acting perpendicular to the rotor disc and stator plate surfaces—exciting disc bending modes and producing a distinct tonal NVH character at harmonics of the pole-pair passing frequency that is not well predicted by radial force-based simulation workflows developed for radial flux machines [52,53,54,55]. The absence of an enclosed cylindrical stator housing also removes the primary acoustic radiation barrier present in radial flux designs, increasing the airborne radiation efficiency of the stator assembly. New coupled electromagnetic–structural–acoustic simulation workflows specifically validated for axial flux geometries are therefore needed before production-ready NVH prediction capability can be claimed for this motor topology. Furthermore, the materials science dimension of NVH is evolving beyond conventional composites: emerging research on nanotechnology-enabled acoustic materials—including carbon nanotube-reinforced polymer composites, graphene-based nano-laminates, and aerogel-nano-composite absorbers—demonstrates specific damping coefficients and sound absorption coefficients substantially exceeding conventional materials at equivalent or lower areal mass [254,255]. Integrating these nano-composite acoustic treatments into BIW and acoustic package designs represents an open research frontier with significant potential for mass-neutral NVH performance improvement. Addressing these gaps through continued interdisciplinary collaboration across acoustics, structural mechanics, electrical engineering, materials science, data science, and psychoacoustics will define the next chapter of automotive NVH research.