Beyond Virtual Gearboxes: From Manufacturing-Aware Simulation to Evidence-Validated Production NVH Digital Twins

1. Introduction

Electrified drivetrains have fundamentally changed the acoustic environment of road vehicles. Reduced combustion-engine masking, high rotational speeds, and increasing structural integration make comparatively low-amplitude periodic excitation from gears, bearings, electric machines, and power electronics more perceptually relevant. At the same time, gearbox simulation has progressed from nominal transmission-error calculations toward measured-geometry contact models, flexible multibody dynamics, vibroacoustics, and perceptual prediction. [1,2,3,8,9]

Manufacturing and assembly variability are now explicitly represented through deterministic deviations, uncertainty propagation, robust design, and surrogate modeling, while virtual-EOL strategies have emerged for electric-drive applications. [4,5,6,7]

Production-oriented evidence extends further downstream: EOL response has been related to vehicle-level sound quality, gearbox digital-twin models can be updated using physical data, and industrial architectures already combine production metrology, model-based prediction, and feedback. [10,11,12,13,14,15,16]

G -> TE -> F_mesh -> F_bearing -> v_housing -> p_acoustic

In this source-to-response chain, G denotes the relevant gear geometry, TE the transmission error, F_mesh the dynamic gear-mesh excitation, F_bearing the force transmitted through the bearing interfaces, v_housing the structural response, and p_acoustic the resulting acoustic field.

X_real = X_nom + ΔX_manufacturing + ΔX_assembly

Y_NVH,i = f(X_real,i, q_i)

Nominally identical drive units can therefore exhibit different transmission-error spectra, bearing forces, resonance amplitudes, housing vibration, radiated sound, and perceived tonal quality. The relevant research question has consequently changed. It is no longer simply whether manufacturing variability can be represented or whether a virtual EOL model can be constructed; it is whether increasingly integrated production-NVH architectures are supported by evidence showing that they reproduce, explain, and ultimately improve the behavior of real manufactured units and production populations.

1.1. Relationship to Existing Reviews

Recent reviews have already examined manufacturing and mounting uncertainties, micro- and macrogeometry, optimization, and robustness in spur and helical gear NVH. The present work therefore does not repeat a general survey of tolerance effects. Instead, it shifts the unit of analysis from deviation-to-response toward manufactured-unit-to-digital-representation-to-validation-evidence, and expands the scope to production metrology, virtual EOL, unit traceability, digital-twin updating, patent activity, manufacturing feedback, population validation, and order-domain traceability. [17]

deviation -> gear response

manufactured unit -> digital representation -> validation evidence

1.2. Objectives and Research Questions

• RQ1. Which manufacturing-aware gearbox NVH capabilities have been conceptually proposed, numerically demonstrated, experimentally validated, and validated on physical production populations?
• RQ2. How does architectural integration maturity compare with the maturity of the corresponding experimental evidence?
• RQ3. Which validation stages remain weak between manufacturing-state information and unit-specific EOL prediction?
• RQ4. Can a common order-domain representation provide traceability between manufacturing processes, tooth-surface geometry, simulated excitation, and measured EOL response?
• RQ5. What evidence is required before a virtual EOL model can credibly replace or reduce physical EOL testing?
• RQ6. Which research directions are required to progress from integrated digital-twin architecture toward evidence-validated production NVH twins?

2. Review Methodology

2.1. Review Approach

This work is structured as a targeted critical review and research perspective rather than a formal systematic review or meta-analysis. The objective is not publication-frequency estimation. Representative evidence was selected to identify capabilities that are already demonstrated, claims that remain insufficiently supported, and combinations of capabilities that appear substantially less mature than their individual components. Recent literature from approximately 2021-2026 received particular attention, while earlier sources were retained when they demonstrated capabilities that might otherwise incorrectly be presented as new.

2.2. Evidence Classes

Four source classes were included: peer-reviewed scientific literature, doctoral dissertations, industrial technical evidence, and patent literature. Peer-reviewed work forms the principal scientific evidential basis. Doctoral dissertations were included where they provided unusually detailed integrated methods or validation. Industrial sources were used to assess implementation maturity such as all-tooth inspection, waviness/order analysis, process monitoring, and manufacturing correction. Patent disclosure was interpreted as architectural evidence, not as proof of predictive accuracy.

2.3. Evidence Taxonomy

Table 1. Evidence coding taxonomy used for the targeted source set.

Code	Capability
MG	Measured physical geometry
TR	Tooth-resolved representation
PD	Real production data
UT	Unit/part traceability
TP	Tolerance population/uncertainty quantification
CP	Gear-contact physics/TE
SD	System dynamics
FS	Flexible structures
VS	Variable-speed analysis
AC	Acoustic prediction
EO	End-of-line relationship
PV	Perceptual/psychoacoustic endpoint
SU	Surrogate or machine-learning model
DT	Digital-twin/model updating
MF	Manufacturing feedback
EV	Experimental validation

For the detailed evidence database, each capability can be coded as 0 = absent, 1 = partial or indirect, and 2 = explicitly demonstrated. A separate primary Architecture and Evidence level was assigned to each coded source for the occupancy analysis.

Table 1. Evidence coding taxonomy used for the targeted source set.

Code	Capability
MG	Measured physical geometry
TR	Tooth-resolved representation
PD	Real production data
UT	Unit/part traceability
TP	Tolerance population/uncertainty quantification
CP	Gear-contact physics/TE
SD	System dynamics
FS	Flexible structures
VS	Variable-speed analysis
AC	Acoustic prediction
EO	End-of-line relationship
PV	Perceptual/psychoacoustic endpoint
SU	Surrogate or machine-learning model
DT	Digital-twin/model updating
MF	Manufacturing feedback
EV	Experimental validation

For the detailed evidence database, each capability can be coded as 0 = absent, 1 = partial or indirect, and 2 = explicitly demonstrated. A separate primary Architecture and Evidence level was assigned to each coded source for the occupancy analysis.

2.4. Architecture Maturity

A0 - Nominal deterministic simulation

X_nom -> Y

Nominal design is evaluated without explicit production variation.

A1 - Variability-aware design

p_assumed(X) -> p(Y)

Manufacturing or assembly deviations are represented statistically.

A2 - As-built simulation

X_measured,i -> Y_i

Measured geometry or physical state of a particular manufactured component is introduced.

A3 - Virtual EOL

X_i -> Y_virtualEOL,i

High-fidelity or surrogate models predict quality-relevant quantities before or instead of selected physical EOL testing.

A4 - Unit-specific NVH digital twin

Manufacturing_i + Geometry_i + Assembly_i + Model_i -> Y_EOL,pred,i

The model becomes explicitly tied to the manufactured unit.

A5 - Closed-loop manufacturing integration

Prediction -> Attribution -> Manufacturing correction

Prediction becomes actionable and feeds back to production.

2.5. Evidence Maturity

E0 - Conceptual evidence. Architecture, patent, workflow, or simulation architecture without direct physical validation of the relevant claim.

E1 - Component validation. A local physical quantity such as contact pattern or TE is experimentally validated.

E2 - System validation. Intermediate or final system quantities such as bearing-interface response, housing vibration, or SPL are compared with measurements.

E3 - Same-unit production validation. The physical unit represented by the model is explicitly the same unit measured downstream.

E4 - Production-population validation. Unit-specific and distributional agreement are evaluated over a manufactured population.

E5 - Intervention validation. A model-selected manufacturing action is implemented and shown experimentally to reduce physical NVH risk.

M = (A, E)

2.6. Interpretation of Empty Cells

The targeted evidence set is not an exhaustive bibliometric census. Accordingly, an empty Architecture x Evidence cell means that no sufficiently explicit example was identified in the reviewed public evidence set. It does not prove that the capability does not exist in proprietary industrial practice.

3. State of the Art: Capabilities That Are Already Established or Emerging

3.1. Manufacturing-Aware Contact Simulation and Robust Design

Manufacturing and assembly deviations are represented using deterministic and probabilistic methods, including global sensitivity analysis, production-aware tolerance optimization, Monte Carlo methods, Latin Hypercube sampling, DoE, ANOVA, and surrogate response surfaces. The evidence therefore shows that uncertainty propagation itself is not the principal unresolved problem. [4,5,6,18,19]

p_assumed(X) ?≈ p_production(X)

The more demanding question is whether the assumed joint distribution is representative of the real manufactured population and whether its propagated NVH distribution matches physical production scatter.

3.2. Measured Geometry and Tooth Individuality

Measured tooth geometry has been introduced into LTCA, three-dimensional contact models, double-flank-derived analytical representations, hypoid-gear simulations, and measured-error transmission-error models. Some approaches average measured profiles, whereas others preserve substantially more tooth-specific or three-dimensional information. [1,2,35,45,46]

Tooth-resolved representation is therefore not a novelty claim by itself. The stronger research problem is production-scaled, NVH-preserving representation of dense geometry. Surrogate work based on generalized topography deviations and controlled tooth-to-tooth microgeometry scattering further demonstrates that the spatial structure of deviations can be deliberately parameterized. [6,20]

3.3. Flexible System Dynamics and Acoustics

Validated elastic multibody and system-level models increasingly include flexible gears, shafts, bearings, housings, variable-speed excitation, and acoustic radiation. Recent work also combines full-system NVH optimization, physics-informed acoustic surrogates, prototype validation, and perceptual assessment. [3,8,21,22,23,32]

The excitation-transfer-radiation chain is therefore no longer a purely conceptual roadmap. The remaining challenge is to repeat this chain for real production units while retaining correspondence between the measured input state and the downstream physical response.

3.4. Virtual EOL

High-fidelity virtual EOL has been demonstrated using transient electric-drive models, large DoE campaigns, high-performance computing, and surrogate models. A separate virtual-EOL framework uses measured topographies of the gears intended for assembly, a full-system elastic multibody model, and a neural-network surrogate to predict acoustic behavior before final assembly. [7,47]

These studies demonstrate that virtual EOL itself is an existing capability. The unresolved issue is the depth of physical same-unit and production-population validation required before virtual predictions can replace or reduce physical quality gates.

3.5. Production Metrology and Order-Domain Manufacturing Control

Industrial production technology has progressed toward all-tooth inspection, advanced waviness/order analysis, ghost-order identification, and closed-loop correction. Patent families additionally describe manufacturing-axis data transformed into predicted EOL orders, rolling-test and manufacturing data expressed in common order coordinates, AI-assisted order attribution, selective NVH-oriented metrology, and production tolerance definition from EOL good/bad populations. [16,26,27,28,29,30,31]

These sources show that empirical manufacturing-to-order traceability is already an industrial reality in selected implementations. The scientific opportunity is to fill the intermediate causal physics between surface deviation, transmission error, transmitted force, and system response.

3.6. Digital-Twin Updating and Data-Driven Adaptation

Gearbox digital-twin research already includes physical-virtual data fusion, parameter updating, domain adaptation, graph-based real-time surrogates, and mechanism-plus-data adaptive residual correction. These capabilities are strongest in fault diagnosis, health monitoring, and structural-response prediction rather than manufacturing-induced production NVH. [12,13,14,33,34]

The gap is therefore an application-integration gap rather than a generic absence of digital-twin model updating.

3.7. Wear and Lifecycle Evolution

Geometry-dependent NVH does not end at the manufacturing state. Recent work explicitly couples three-dimensional tooth-surface geometry with wear and nonlinear dynamics, while inhomogeneous wear models show that spatially non-uniform degradation can alter vibration behavior. [24,25]

G = G(t), NVH = NVH[G(t)]

Lifecycle modeling is an important extension, but the immediate production problem remains the credible prediction of G0 -> NVH0 across real manufactured populations.

4. Results

4.1. Architecture x Evidence Occupancy

The coded evidence set contains 46 external sources. The contextual review in Ref. [17] was used to position the present contribution relative to existing review literature but was intentionally excluded from the occupancy counts.

The largest concentration occurs around A1-A3 and E1-E2, corresponding to manufacturing-aware physical simulation, as-built models, system dynamics, acoustics, and component/system validation. Highly integrated architecture is also visible: 13 of the 46 coded sources were assigned to A4-A5.

N = 46

N(A >= 4) = 13 (28.3%)

N(A >= 4 and E >= 3) = 0

No source in the targeted evidence set simultaneously reached high architectural integration and same-unit production, population, or intervention-level validation according to the adopted definitions. This is the principal quantitative observation of the review.

Figure 1. Architecture x Evidence occupancy map derived from the 46-source targeted evidence set. Architectural integration and validation maturity were coded independently. The outlined A4-A5/E3-E5 region represents the evidence-validation frontier: highly integrated architectures are increasingly represented, whereas same-unit production, population-level, and intervention validation remain sparsely populated in the reviewed public evidence set.

Figure 1. Architecture x Evidence occupancy map derived from the 46-source targeted evidence set. Architectural integration and validation maturity were coded independently. The outlined A4-A5/E3-E5 region represents the evidence-validation frontier: highly integrated architectures are increasingly represented, whereas same-unit production, population-level, and intervention validation remain sparsely populated in the reviewed public evidence set.

4.2. Architecture-Evidence Asymmetry

The empty upper-right region should not be interpreted as the absence of advanced production digital-twin architectures. On the contrary, recent patents and industrial implementations describe unit-specific data, acoustic digital twins, virtual EOL, pass/fail decisions, order-based process attribution, and manufacturing adjustment. The asymmetry instead indicates that architectural integration is advancing faster than publicly demonstrated production-level validation. [15,16,26,27,28,29,30,31]

Architecture maturity > public evidence maturity (high-integration region)

4.3. Claim-Evidence Gradient

The evidence synthesis also reveals a systematic decline in support as claims move from conceptual/simulation capability toward same-unit validation, population validation, and intervention validation. Manufacturing variability-to-excitation and measured geometry-to-TE are strongly supported, whereas population NVH prediction, unit-specific acoustic twins, virtual replacement of physical EOL, and closed-loop intervention validation have substantially thinner public evidence.

Figure 2. Claim-Evidence maturity synthesis. Cells qualitatively summarize the strength of evidence identified for each technical claim across successive validation levels, from conceptual or patent disclosure to intervention validation. The figure is an evidence synthesis rather than a bibliometric frequency map.

Figure 2. Claim-Evidence maturity synthesis. Cells qualitatively summarize the strength of evidence identified for each technical claim across successive validation levels, from conceptual or patent disclosure to intervention validation. The figure is an evidence synthesis rather than a bibliometric frequency map.

4.4. Order-Domain Traceability

Manufacturing, metrology, rolling-test, and EOL data can already share common order-domain information in industrial practice. However, empirical order correspondence does not by itself establish the causal physical sequence from measured surface deviation through transmission error and transmitted forces to structural and acoustic response. [16,26,27,28,29,30,31]

O_process -> O_surface -> O_TE -> O_force -> O_housing -> O_EOL

Figure 3 proposes a gear-specific production digital thread in which physical and spectral traceability coexist, while model-updating and manufacturing-correction loops remain distinct.

4.5. Evidence from High-Speed, Perceptual, and Production Datasets

Additional evidence confirms that several adjacent capabilities are already mature enough that they should not be framed generically as future work. Excitation-oriented tolerancing has been addressed in dedicated doctoral research; stepped planetary stages have been optimized under manufacturing and assembly deviations with test-rig validation; high-speed electromechanical gearbox dynamics have been investigated experimentally; optimized gear NVH has been validated beyond measurement uncertainty; and real industrial EV-reducer datasets now provide multi-machine, multi-batch manufacturing-quality information. [36,37,38,39,40,41,42,43,44]

These results further reinforce the central conclusion: the opportunity lies in integrating mature sub-capabilities with deeper production-unit and population evidence, not in presenting those sub-capabilities individually as novel.

5. Discussion

5.1. From Technology Gaps to Evidence Gaps

The review shows that many previously proposed technology gaps have become outdated. Manufacturing-aware contact simulation, measured geometry, tooth-resolved modeling, UQ, flexible drivetrain NVH, acoustic prediction, surrogate modeling, virtual EOL, digital-twin updating, production metrology, and manufacturing feedback are all represented in contemporary research or industrial development.

integrated capability -> validated capability

The more defensible frontier is therefore evidence depth rather than the mere existence of additional model components.

5.2. Predictive Credibility and Four Traceability Requirements

C_prediction = f(F_model, F_input, T_part, T_physical, T_spectral, F_statistical, D_validation)

Here, F_model represents numerical fidelity, F_input the fidelity of physical inputs, T_part part traceability, T_physical physical causality, T_spectral order-domain correspondence, F_statistical population fidelity, and D_validation validation depth. A highly detailed numerical model cannot indefinitely compensate for weak production grounding.

Part traceability. All relevant data belong to the same physical unit.

Part_i: Process_i -> Geometry_i -> Assembly_i -> EOL_i

Physical traceability. The intermediate causal mechanism is retained.

Process -> Geometry -> Contact -> Force -> Response

Spectral traceability. Periodic features are followed through a common order-domain representation.

O_process -> O_geometry -> O_TE -> O_force -> O_EOL

Statistical traceability. The virtual population corresponds to the physical population.

p(X_model) ≈ p(X_production), p(Y_model) ≈ p(Y_production)

5.3. Assumed Versus Measured Production Distributions

Many uncertainty studies use statistically independent variables. Real manufacturing can create covariance through common tooling, tool wear, machine drift, thermal history, batch, fixture state, dressing condition, and assembly sequence.

p_production(x1,...,xn) != Π p(xj)

A high-value testable question is whether measured joint production distributions reproduce EOL scatter more accurately than conventional independently sampled drawing tolerances.

5.4. Excitation Versus Transfer-Path Variability

Y_i(ω) = H_i(ω) F_i(ω)

Var(Y) = Var_F + Var_H + Var_interaction + Var_measurement

The F-side contains gear excitation effects such as tooth geometry, pitch, runout, waviness, and contact state. The H-side contains bearing preload and clearance, shaft alignment, housing properties, assembly state, and mounting conditions. Identical EOL symptoms can therefore require fundamentally different corrective actions depending on whether variability originates in excitation or the transfer path.

5.5. NVH-Preserving Geometry Reduction

Δg(θ,u,v) -> z_NVH

dim(z_NVH) << dim(Δg), Y(z_NVH) ≈ Y(Δg)

Production-scale use of complete tooth-surface metrology requires a reduced representation that retains mesh harmonics, tooth-to-tooth modulation, waviness, runout, localized deviations, sidebands, and ghost-order-generating components. This is more demanding than generic dimensionality reduction.

5.6. Physics-Preserving Surrogates

X -> Y(O,n,T)

Scalar metrics such as R², RMSE, or MAE can hide errors in spectral mechanism. A production NVH surrogate should preserve dominant order, order amplitude, sideband spacing, resonance crossing, and order-dependent scatter.

ΔO_peak, ΔA_O, Δn_res, Δσ_O

These physics-preserving metrics should complement conventional regression accuracy.

5.7. Virtual EOL as a Quality-Control Problem

If virtual EOL is intended to reduce or replace a physical quality-control test, the relevant validation problem is ultimately classification rather than regression alone.

Sensitivity, Specificity, False Reject Rate, False Pass Rate

False pass is particularly critical because it permits an acoustically non-conforming physical unit to be accepted. An impressive R² for a simulation surrogate is therefore insufficient evidence that a physical EOL quality gate can safely be removed.

5.8. Minimum Validation Protocol

Gate 1 - Geometry/contact. Validate measured geometry, contact pattern, and STE/DTE.

Geometry_i -> TE_i

Gate 2 - Interface force. Validate transmitted force or a suitable interface-response proxy.

TE_i -> F_bearing,i

Gate 3 - Structural response. Validate order amplitudes and resonance crossings.

F_i -> a_housing,i

Gate 4 - EOL response. Compare the same physical units.

a_housing,i -> Y_EOL,i

Gate 5 - Population. Evaluate μ_O, σ_O, P95,O, P(NOK), and relevant covariance.

i = 1,...,N

5.9. Controlled Variability and Acoustic Quality

Targeted tooth-to-tooth microgeometry scattering demonstrates that manufacturing variation can also be deliberately engineered to redistribute spectral energy. Therefore, minimum geometric deviation or minimum peak-to-peak TE does not automatically imply minimum tonality. [20]

min TE_pp != necessarily min Tonality

The distinction between uncontrolled manufacturing variability and designed acoustic variability should be preserved. Future optimization may therefore combine tonality, loudness, variance, manufacturing cost, durability, and efficiency.

5.10. Lifecycle Extension

G_0 -> NVH_0 -> Wear -> G_1 -> NVH_1

Wear-related studies show that surface geometry and dynamic behavior evolve during service. A lifecycle twin can extend the same framework, but the production-state problem G0 -> NVH0 should first be demonstrated credibly over physical production populations. [24,25]

6. Priority Research Agenda

Priority 1 - Same-unit digital thread. Process_i -> Geometry_i -> Simulation_i -> EOL_i.

Priority 2 - Population replication. Repeat the same-unit chain for i = 1,...,N.

Priority 3 - Measured production distributions. Compare p_assumed(X) with p_production(X).

Priority 4 - Variance decomposition. Separate Var_F from Var_H and interaction effects.

Priority 5 - NVH-preserving metrology reduction. Convert dense all-tooth data into simulation-ready variables without losing order topology.

Priority 6 - Order-preserving surrogates. Predict complete variable-speed order maps rather than scalar KPIs only.

Priority 7 - Production-quality virtual-EOL validation. Quantify false-pass and false-reject rates.

Priority 8 - Physical order traceability. Validate O_process -> O_surface -> O_TE -> O_force -> O_EOL.

Priority 9 - Intervention validation. Test whether model-selected manufacturing corrections reduce measured NVH.

Priority 10 - Lifecycle and perceptual extension. Extend the validated production framework toward wear and customer-relevant sound quality.

7. Limitations

The quantitative evidence map requires careful interpretation. First, the 46-source set is a targeted critical evidence sample rather than an exhaustive systematic bibliometric database; cell occupancy therefore should not be interpreted as publication prevalence. Second, heterogeneous source types were intentionally included, and patents or industrial technical evidence do not provide the same scientific proof as peer-reviewed experiments. Third, assigning one primary (A,E) coordinate reduces multidimensional studies to their dominant claim. Fourth, proprietary industrial implementations may possess substantially greater validation than is publicly disclosed. The empty high-integration/high-evidence cells should therefore be read as limited publicly identified evidence, not as proven technological absence.

8. Conclusions

Gearbox NVH simulation has advanced substantially beyond nominal deterministic design. Manufacturing-aware contact simulation, measured tooth geometry, tooth-resolved representations, uncertainty propagation, flexible drivetrain dynamics, acoustic prediction, surrogate modeling, virtual EOL, digital-twin updating, production metrology, and manufacturing feedback are all represented in contemporary research or industrial development.

The principal research gap is therefore no longer adequately described as a lack of manufacturing awareness or digital-twin architecture. Instead, the evidence points toward an architecture-evidence gap.

N(A >= 4) = 13

N(A >= 4 and E >= 3) = 0

Within the targeted 46-source coded set, highly integrated architectures are visible, but corresponding public evidence for same-unit production validation, production-population validation, and controlled intervention remains substantially weaker.

Part + Physical + Spectral + Statistical traceability

Architecturally capable twin -> Evidence-validated production twin

Future studies should therefore prioritize measured production distributions, unit-specific validation, excitation-versus-transfer-path variance decomposition, NVH-preserving geometry reduction, order-preserving surrogate models, virtual-EOL quality-control metrics, and experimentally verified manufacturing interventions. The next generation of gearbox NVH simulation will be defined less by how precisely an ideal gearbox can be represented and more by how convincingly the digital model can be shown to represent, explain, and ultimately improve the acoustic behavior of real manufactured populations.

Table 2. Coded Evidence Matrix. The 46 coded sources used for the Architecture x Evidence occupancy analysis are listed below. Reference [17] is the contextual comparison review and is not part of the 46-source coded set.

Ref.	Source/title	Year	Type	A	E	Evidence codes
[1]	Im, D.. Analysis of gear transmission error in helical gear using enhanced tooth contact analysis model considering measured tooth profile errors	2025	Peer-reviewed	A2	E1	MG, CP, EV
[2]	Grünberger, J.. Multibody simulation of spur gears using 3D contact modeling and real-world tooth geometry	2026	Peer-reviewed	A2	E2	MG, TR, CP, SD, EV
[3]	Wischmann, S.. Validation of models for calculating the NVH behavior of gearbox systems in an elastic multibody simulation	2025	Peer-reviewed	A2	E2	CP, SD, FS, VS, AC, EV
[4]	Oberneder, F.. The impact of manufacturing deviations on gearbox performance using a global sensitivity analysis	2026	Peer-reviewed	A1	E1	TP, CP, SU
[5]	Roth, L.. Design of profile corrections and tolerances in cylindrical gears for noise-sensitive applications in line with production efforts	2025	Peer-reviewed	A1	E1	TP, CP, SU
[6]	Willecke, M.. Surrogate model based prediction of transmission error characteristics based on generalized topography deviations	2023	Peer-reviewed	A2	E1	TR, CP, SU
[7]	Klarin, M.. Understanding End-of-Line NVH Scatter Bands in Electric Drives with High-Precision Simulations	2025	SAE technical paper	A3	E1	TP, CP, SD, FS, VS, SU, EO
[8]	Li, S.. A new method for analyzing gearbox vibration and noise in electric drive systems based on different modal forms	2026	Peer-reviewed	A3	E2	CP, SD, FS, VS, AC, EV
[9]	Souza, T.. The use of an artificial neural network for assessing tone perception in electric powertrain NVH	2024	Peer-reviewed	A3	E2	SD, FS, AC, PV, SU, EV
[10]	Su, C.. Correlation of the sound quality and vibration of end of line testing for automatic transmission	2024	Peer-reviewed	A3	E4	PD, UT, EO, PV, EV
[11]	Arvanitis, A.. Predicting Electric Vehicle Interior Noise Quality from Drive Unit End-of-Line Vibration Testing	2026	SAE technical paper	A3	E3	PD, UT, EO, PV, EV
[12]	Xia, J.. Digital twin-assisted gearbox dynamic model updating toward fault diagnosis	2023	Peer-reviewed	A4	E2	SD, DT, SU, EV
[13]	Xia, J.. A novel digital twin-driven approach based on physical-virtual data fusion for gearbox fault diagnosis	2023	Peer-reviewed	A4	E2	SD, DT, SU, EV
[14]	Zhu, P.. Digital twin-enabled entropy regularized wavelet attention domain adaptation network for gearboxes fault diagnosis without fault data	2025	Peer-reviewed	A4	E2	SD, DT, SU, EV
[15]	Magna International Inc.. Digital twin for manufacturing	2026	Patent	A5	E0	PD, UT, DT, AC, MF
[16]	Turich, A.. E-Drive Gear Inspection and Noise Analysis	2022	Industrial technical evidence	A5	E2	MG, TR, PD, MF
[18]	De Smet, B.. Reducing Tolerance-Induced Spread on Transmission Error in Planetary Gear Stages Using Monte Carlo Simulations and Surrogate Models Acquired From a Design of Experiments Approach	2025	Peer-reviewed	A1	E1	TP, CP, SU
[19]	Najib, R.. Assessing the impact of manufacturing uncertainties on the static and dynamic response of spur gear pairs	2024	Peer-reviewed	A1	E2	TP, CP, EV
[20]	Mann, A.. Targeted micro geometry scattering for NVH optimization of cylindrical gears in continuous generating grinding	2025	Peer-reviewed	A2	E1	TR, CP, MF
[21]	Cheng, et al.. Towards precise and rapid gearbox noise prediction: a physics-informed region-aware acoustic predictor	2026	Peer-reviewed	A3	E2	AC, SU, DT, EV
[22]	Cianciotta, L.. Multi-Objective Optimization of Gear Design of E-Axles to Improve Noise Emission and Load Distribution	2025	Peer-reviewed	A3	E2	CP, SD, FS, AC, EV
[23]	He, et al.. Optimization of Lightweight Gear Blanks to Improve NVH for an Electric Drive Unit	2026	SAE technical paper	A3	E2	CP, SD, FS, VS, AC, PV, EV
[24]	Liu, C.. Influence of Tooth Surface Wear on Dynamic Characteristics of Gear Systems Considering Three-Dimensional Tooth Surface Geometric Errors	2025	Peer-reviewed	A2	E1	MG, CP, SD
[25]	Brethee, K. F.. Nonlinear Characterisation of Wind Turbine Gearbox Vibration Dynamics Driven by Inhomogeneous Helical Gear Wear	2026	Peer-reviewed	A2	E1	CP, SD, VS
[26]	Reishauer AG. Method of monitoring the condition of a gear cutting machine	2023	Patent	A5	E0	PD, UT, EO, MF
[27]	Klingelnberg AG. Analysis of gear cutting process by means of roller testing	2025	Patent	A5	E0	PD, UT, EO, MF
[28]	Klingelnberg AG. Method and manufacturing system	2026	Patent	A5	E0	MG, TR, PD, UT, MF
[29]	Klingelnberg AG. Method and device for order analysis	2026	Patent	A5	E0	PD, EO, SU, MF
[30]	KISSsoft AG. Method for determining the expected noise emissions of a gear pair	2025	Patent	A4	E0	MG, TR, CP, AC
[31]	Klingelnberg AG. Device for rolling testing for spur wheels	2025	Patent	A5	E0	PD, UT, EO, MF
[32]	Mughal, H.. Comprehensive tribodynamic analysis of lubricated gear contacts for direct identification of gear whine noise in electric drive units	2025	Peer-reviewed	A2	E2	CP, SD, AC, EV
[33]	Zhu, B.. Digital twin surrogate modeling for real-time monitoring of gear transmissions using a dynamic graph attention network	2026	Peer-reviewed	A4	E2	SD, SU, DT, EV
[34]	Zhang, P.. A mechanism and data-driven adaptive update framework for digital twin model of electric drive assembly gear transmission system	2026	Peer-reviewed	A4	E2	SD, SU, DT, EV
[35]	Wang, et al.. A Novel Method for Obtaining Analytical Parameters Based on Double-Flank Measurement	2024	Peer-reviewed	A2	E2	MG, PD, EV
[36]	Nohl, et al.. Quasi-static modeling of long-wave axis deviations in planetary gears for transmission error signal analysis	2025	Peer-reviewed	A1	E1	TP, CP
[37]	Ahmad, M.. Anregungsgerechte Auslegung und Tolerierung von schnelllaufenden Getrieben unter Berücksichtigung kurz- und langwelliger Verzahnungsabweichungen	2023	Dissertation	A1	E1	TR, TP, CP
[38]	Westphal, C.. Gear Design for Stepped Planetary Gear Stages	2026	Dissertation	A2	E2	TP, CP, SD, FS, EV
[39]	Schweigert, C.. High-Speed Gearboxes for Electromechanical Drivetrains	2025	Dissertation	A2	E2	SD, VS, EV
[40]	Mabrouk, M.. Optimizing gear NVH performance using genetic algorithm-based on tooth profile modification and a comparative analysis of simulation and experimental outcomes	2026	Peer-reviewed	A2	E2	CP, AC, EV
[41]	Xie, A.. Multi-objective micro-geometry optimization of high-speed e-drive gears for NVH and strength	2026	Peer-reviewed	A2	E2	CP, VS, EV
[42]	Ge, et al.. Multifield coupled dynamics modeling and vibration characteristics of electric-vehicle electric drive systems	2024	Peer-reviewed	A3	E1	CP, SD, VS
[43]	Souza, T.. Design of electric powertrains to achieve NVH performance using autoencoders and a physical meaningful latent space	2025	Peer-reviewed	A3	E1	AC, PV, SU
[44]	Guo, et al.. A dynamic meshing transmission dataset for manufacturing quality inspection of electric vehicle reducer gears	2026	Peer-reviewed data descriptor	A2	E2	PD, VS, SU, EV
[45]	Liang, et al.. Modelling method, simulation and experimental verification of hypoid gear involved tooth surface deviation under manufacturing process	2023	Peer-reviewed	A2	E1	MG, CP, EV
[46]	Zhang, et al.. Transmission error modeling and analysis of herringbone gear drive based on the measured manufacturing errors	2025	Peer-reviewed	A2	E1	MG, CP, EV
[47]	Willecke, M.. Virtual End of Line Test: Prediction of the Acoustic Behavior of Gearboxes Based on Topographic Deviations Using Neural Networks	2023	Conference technical paper	A3	E0	MG, TR, CP, SD, SU, EO

Table 2. Coded Evidence Matrix. The 46 coded sources used for the Architecture x Evidence occupancy analysis are listed below. Reference [17] is the contextual comparison review and is not part of the 46-source coded set.

Ref.	Source/title	Year	Type	A	E	Evidence codes
[1]	Im, D.. Analysis of gear transmission error in helical gear using enhanced tooth contact analysis model considering measured tooth profile errors	2025	Peer-reviewed	A2	E1	MG, CP, EV
[2]	Grünberger, J.. Multibody simulation of spur gears using 3D contact modeling and real-world tooth geometry	2026	Peer-reviewed	A2	E2	MG, TR, CP, SD, EV
[3]	Wischmann, S.. Validation of models for calculating the NVH behavior of gearbox systems in an elastic multibody simulation	2025	Peer-reviewed	A2	E2	CP, SD, FS, VS, AC, EV
[4]	Oberneder, F.. The impact of manufacturing deviations on gearbox performance using a global sensitivity analysis	2026	Peer-reviewed	A1	E1	TP, CP, SU
[5]	Roth, L.. Design of profile corrections and tolerances in cylindrical gears for noise-sensitive applications in line with production efforts	2025	Peer-reviewed	A1	E1	TP, CP, SU
[6]	Willecke, M.. Surrogate model based prediction of transmission error characteristics based on generalized topography deviations	2023	Peer-reviewed	A2	E1	TR, CP, SU
[7]	Klarin, M.. Understanding End-of-Line NVH Scatter Bands in Electric Drives with High-Precision Simulations	2025	SAE technical paper	A3	E1	TP, CP, SD, FS, VS, SU, EO
[8]	Li, S.. A new method for analyzing gearbox vibration and noise in electric drive systems based on different modal forms	2026	Peer-reviewed	A3	E2	CP, SD, FS, VS, AC, EV
[9]	Souza, T.. The use of an artificial neural network for assessing tone perception in electric powertrain NVH	2024	Peer-reviewed	A3	E2	SD, FS, AC, PV, SU, EV
[10]	Su, C.. Correlation of the sound quality and vibration of end of line testing for automatic transmission	2024	Peer-reviewed	A3	E4	PD, UT, EO, PV, EV
[11]	Arvanitis, A.. Predicting Electric Vehicle Interior Noise Quality from Drive Unit End-of-Line Vibration Testing	2026	SAE technical paper	A3	E3	PD, UT, EO, PV, EV
[12]	Xia, J.. Digital twin-assisted gearbox dynamic model updating toward fault diagnosis	2023	Peer-reviewed	A4	E2	SD, DT, SU, EV
[13]	Xia, J.. A novel digital twin-driven approach based on physical-virtual data fusion for gearbox fault diagnosis	2023	Peer-reviewed	A4	E2	SD, DT, SU, EV
[14]	Zhu, P.. Digital twin-enabled entropy regularized wavelet attention domain adaptation network for gearboxes fault diagnosis without fault data	2025	Peer-reviewed	A4	E2	SD, DT, SU, EV
[15]	Magna International Inc.. Digital twin for manufacturing	2026	Patent	A5	E0	PD, UT, DT, AC, MF
[16]	Turich, A.. E-Drive Gear Inspection and Noise Analysis	2022	Industrial technical evidence	A5	E2	MG, TR, PD, MF
[18]	De Smet, B.. Reducing Tolerance-Induced Spread on Transmission Error in Planetary Gear Stages Using Monte Carlo Simulations and Surrogate Models Acquired From a Design of Experiments Approach	2025	Peer-reviewed	A1	E1	TP, CP, SU
[19]	Najib, R.. Assessing the impact of manufacturing uncertainties on the static and dynamic response of spur gear pairs	2024	Peer-reviewed	A1	E2	TP, CP, EV
[20]	Mann, A.. Targeted micro geometry scattering for NVH optimization of cylindrical gears in continuous generating grinding	2025	Peer-reviewed	A2	E1	TR, CP, MF
[21]	Cheng, et al.. Towards precise and rapid gearbox noise prediction: a physics-informed region-aware acoustic predictor	2026	Peer-reviewed	A3	E2	AC, SU, DT, EV
[22]	Cianciotta, L.. Multi-Objective Optimization of Gear Design of E-Axles to Improve Noise Emission and Load Distribution	2025	Peer-reviewed	A3	E2	CP, SD, FS, AC, EV
[23]	He, et al.. Optimization of Lightweight Gear Blanks to Improve NVH for an Electric Drive Unit	2026	SAE technical paper	A3	E2	CP, SD, FS, VS, AC, PV, EV
[24]	Liu, C.. Influence of Tooth Surface Wear on Dynamic Characteristics of Gear Systems Considering Three-Dimensional Tooth Surface Geometric Errors	2025	Peer-reviewed	A2	E1	MG, CP, SD
[25]	Brethee, K. F.. Nonlinear Characterisation of Wind Turbine Gearbox Vibration Dynamics Driven by Inhomogeneous Helical Gear Wear	2026	Peer-reviewed	A2	E1	CP, SD, VS
[26]	Reishauer AG. Method of monitoring the condition of a gear cutting machine	2023	Patent	A5	E0	PD, UT, EO, MF
[27]	Klingelnberg AG. Analysis of gear cutting process by means of roller testing	2025	Patent	A5	E0	PD, UT, EO, MF
[28]	Klingelnberg AG. Method and manufacturing system	2026	Patent	A5	E0	MG, TR, PD, UT, MF
[29]	Klingelnberg AG. Method and device for order analysis	2026	Patent	A5	E0	PD, EO, SU, MF
[30]	KISSsoft AG. Method for determining the expected noise emissions of a gear pair	2025	Patent	A4	E0	MG, TR, CP, AC
[31]	Klingelnberg AG. Device for rolling testing for spur wheels	2025	Patent	A5	E0	PD, UT, EO, MF
[32]	Mughal, H.. Comprehensive tribodynamic analysis of lubricated gear contacts for direct identification of gear whine noise in electric drive units	2025	Peer-reviewed	A2	E2	CP, SD, AC, EV
[33]	Zhu, B.. Digital twin surrogate modeling for real-time monitoring of gear transmissions using a dynamic graph attention network	2026	Peer-reviewed	A4	E2	SD, SU, DT, EV
[34]	Zhang, P.. A mechanism and data-driven adaptive update framework for digital twin model of electric drive assembly gear transmission system	2026	Peer-reviewed	A4	E2	SD, SU, DT, EV
[35]	Wang, et al.. A Novel Method for Obtaining Analytical Parameters Based on Double-Flank Measurement	2024	Peer-reviewed	A2	E2	MG, PD, EV
[36]	Nohl, et al.. Quasi-static modeling of long-wave axis deviations in planetary gears for transmission error signal analysis	2025	Peer-reviewed	A1	E1	TP, CP
[37]	Ahmad, M.. Anregungsgerechte Auslegung und Tolerierung von schnelllaufenden Getrieben unter Berücksichtigung kurz- und langwelliger Verzahnungsabweichungen	2023	Dissertation	A1	E1	TR, TP, CP
[38]	Westphal, C.. Gear Design for Stepped Planetary Gear Stages	2026	Dissertation	A2	E2	TP, CP, SD, FS, EV
[39]	Schweigert, C.. High-Speed Gearboxes for Electromechanical Drivetrains	2025	Dissertation	A2	E2	SD, VS, EV
[40]	Mabrouk, M.. Optimizing gear NVH performance using genetic algorithm-based on tooth profile modification and a comparative analysis of simulation and experimental outcomes	2026	Peer-reviewed	A2	E2	CP, AC, EV
[41]	Xie, A.. Multi-objective micro-geometry optimization of high-speed e-drive gears for NVH and strength	2026	Peer-reviewed	A2	E2	CP, VS, EV
[42]	Ge, et al.. Multifield coupled dynamics modeling and vibration characteristics of electric-vehicle electric drive systems	2024	Peer-reviewed	A3	E1	CP, SD, VS
[43]	Souza, T.. Design of electric powertrains to achieve NVH performance using autoencoders and a physical meaningful latent space	2025	Peer-reviewed	A3	E1	AC, PV, SU
[44]	Guo, et al.. A dynamic meshing transmission dataset for manufacturing quality inspection of electric vehicle reducer gears	2026	Peer-reviewed data descriptor	A2	E2	PD, VS, SU, EV
[45]	Liang, et al.. Modelling method, simulation and experimental verification of hypoid gear involved tooth surface deviation under manufacturing process	2023	Peer-reviewed	A2	E1	MG, CP, EV
[46]	Zhang, et al.. Transmission error modeling and analysis of herringbone gear drive based on the measured manufacturing errors	2025	Peer-reviewed	A2	E1	MG, CP, EV
[47]	Willecke, M.. Virtual End of Line Test: Prediction of the Acoustic Behavior of Gearboxes Based on Topographic Deviations Using Neural Networks	2023	Conference technical paper	A3	E0	MG, TR, CP, SD, SU, EO