A digital twin-based fault simulation framework for aircraft electromechanical actuators: method and demonstration

Abstract

Electromechanical actuators have become key components in next-generation aircraft architectures, particularly under More Electric Aircraft and Power-by-Wire paradigms. However, their operational complexity, compounded by mechanical-electrical interactions, introduces failure modes that are both difficult to detect and insufficiently represented in existing datasets. This paper presents a comprehensive digital twin-based framework developed to simulate and analyse EMA behaviour under both nominal and faulty conditions. Implemented using MATLAB Simulink and Simscape, the framework comprises modular voltage and load profiles, structured fault injection mechanisms, and labelled data generation tools. This work investigates the signal responses to various fault types, including mechanical backlash, voltage drop, and electrical resistance anomalies, both in isolation and combination. The simulation output enables systematic feature extraction and evaluation for diagnostics and health indicator development. The framework generated a high-fidelity dataset of 70,000 labelled samples, which demonstrated excellent feature separability for both single and compound faults under Principal Component Analysis. This research addresses the aerospace industry’s pressing need for synthetic, fault-labelled data to train and validate diagnostic algorithms and offers a scalable methodology applicable to diverse actuator configurations. The resulting openable, labelled dataset and modular scripts enable reproducible benchmarking for EMA health monitoring and will be extended to physics-informed prognostics in subsequent work.

1 Introduction

The aviation industry is undergoing a significant transformation driven by the pursuit of greater efficiency, reduced emissions, and enhanced operational reliability. This transition is particularly visible in the migration from traditional hydraulic systems to electromechanical actuators (EMAs) within the framework of More Electric Aircraft (MEA) and Power-by-Wire (PbW) architectures (Zaporozhets, 2020; Emmanouil, 2020; Janson et al., 2017). EMAs offer several advantages: they eliminate the need for hydraulic lines, reduce system weight, simplify maintenance procedures, and integrate seamlessly with digital control systems (MOOG, 2020).

Despite these advantages, EMAs present unique diagnostic challenges. Their failure modes span both electrical and mechanical domains, with interactions between motor characteristics, load dynamics, gear backlash, thermal effects, and sensor degradation (Fu et al., 2017; Wang et al., 2022). Detecting and isolating such faults requires multidomain sensing, cross-signal interpretation, and often, predictive insights into degradation patterns. Traditional data collection methods, dependent on physical testing or operational incident logging, struggle to produce the volume and variety of labelled fault data required to train modern diagnostic and prognostic algorithms.

The Digital Twin (DT) paradigm has emerged as a promising solution to this challenge. A DT is a virtual representation of a physical asset that mirrors its structure, behaviour, and performance (Glaessgen and Stargel, 2012). When applied to aircraft actuators, DTs enable engineers to simulate operational behaviour under nominal and faulty conditions, generate synthetic datasets, evaluate fault progression, and test health management strategies before deployment.

The increasing electrification of aerospace systems parallels broader trends observed in large-scale renewable energy integration, where accurate demand modelling and predictive frameworks are essential for system reliability and planning (Arévalo et al., 2024). As aircraft actuation architectures transition toward fully electric configurations, DT models provide a structured foundation for managing electrical load complexity and supporting predictive health management.

This paper presents a DT-based fault simulation framework specifically designed for aircraft EMAs. The system supports a wide range of simulation scenarios, including multiple fault types and durations, offering a scalable and reproducible method for synthetic health data generation.

2 Failure modes and system-level impacts

2.1 The role of DTs in system health management

The concept of a “twin” originated with NASA’s Apollo program (Glaessgen and Stargel, 2012), where an identical vehicle on Earth was used to mirror and troubleshoot the spacecraft in orbit. The modern DT, first conceptualised by Grieves in 2002 (Grieves, 2016; Grieves, 2023), has evolved into a multi-physics, multi-scale, probabilistic simulation that is continuously updated with data from its physical counterpart. In the context of PHM (Liu et al., 2018), DTs serve two primary roles: as a real-time mirror for operational monitoring and as a simulation platform for data generation. While real-time monitoring is a long-term goal, the immediate challenge for many new systems like EMAs is the lack of historical data needed to build diagnostic models in the first place. Recent literature (Quattrocchi et al., 2021; Cole and Tang, 2022; Wang et al., 2023; Annaz and Kaluarachchi, 2023) recognises this challenge and supports the use of simulation-generated data as a legitimate foundation for DT development following a multiple cycle framework as shown in Figure 1, especially for exploring rare fault modes or for systems where historical failure records are non-existent.

FIGURE 1

2.2 Evolution of aircraft actuation systems

For decades, centralised hydraulic systems were the standard for aircraft flight control, valued for their reliability and power density (Maré and Fu, 2017). However, they suffer from drawbacks including high weight, complex maintenance, and vulnerability to single-point failures in hydraulic lines. The MEA initiative has driven the adoption of electrically powered actuators, which are categorised into two main types: Electro-Hydraulic Actuators (EHAs) and Electro-Mechanical Actuators (EMAs). EHAs are self-contained hydraulic systems driven by an electric motor, offering a bridge between legacy and fully electric architectures. EMAs, however, completely eliminate hydraulic fluid, converting electrical energy directly into mechanical motion via motors and gear trains. This offers the greatest benefits in weight reduction and efficiency but introduces new failure modes and thermal management challenges (Lawson and Pointon, 2008; Wager et al., 2011; Xu et al., 2014).

2.3 EMA components and failure characteristics

EMAs are composed of tightly integrated electrical and mechanical components. The primary subsystems include the electric motor, gear transmission (often screw–nut or planetary mechanisms), electronic control unit (ECU), and position feedback sensors. These subsystems are jointly responsible for converting electrical energy into controlled mechanical motion.

The combined effect of multiple fault modes is both realistic and diagnostically challenging. In scenarios such as FM6 (backlash + resistance + undervoltage), overlapping symptoms can mask the individual fault signatures, requiring multi-feature fusion for accurate classification.

To simulate these conditions faithfully, it is essential to model both the physical fault mechanisms and their time-dependent injection profiles. The next sections introduce the DT design, and the configurable simulation environment used to replicate these effects.

3 DT framework design

The methodology follows a structured engineering approach to systematically develop, operate, and validate the DT for data generation. The methodology follows a structured engineering approach, with the framework implemented in the MATLAB/Simulink and Simscape environment for its multi-domain modelling capabilities and flexibility in script-based automation.

3.1 Case study: A350XWB inboard flap actuator

This work emulates an inboard flap-like duty cycle with A350XWB-inspired loading assumptions for demonstration (AIRBUS, 2017). While the actual aircraft uses a mix of hydraulic and electro-hydraulic systems, its advanced fly-by-wire architecture provides a suitable context for exploring a fully electromechanical solution. Figure 2 shows detailed system structure and energy transmission powertrain of a typical EMA referring to Figure 3. The simulation models a 30-s flap extension cycle, which is representative of a pre-flight check. To ensure fault signatures are observable, the applied mechanical loads of 8, 10, and 12 kN are derived from estimated aerodynamic forces during in-flight deployment, creating a challenging test scenario for the actuator.

FIGURE 2

FIGURE 3

The selected load levels are justified by estimating the aerodynamic force acting on the inboard flap during typical approach conditions. Given:

Flight speed during descending:

Flight speed during approaching:

Flight speed at FULL EXTENSION:

Air density , Flap area during IDLE and Stage 1 , Flap area during Stage 1 and Stage 2 , Flap area during Stage 2 and Stage 3 , Flap area during Stage 3 and FULL EXTENSION , Flap lift coefficient , Then,

Dynamic pressures

Aerodynamic force during IDLE to Stage 1:

Aerodynamic force during Stage 1 to Stage 2:

Aerodynamic force during Stage 2 to Stage 3:

Aerodynamic force during Stage 3 to FULL EXTENSION:

Therefore, in this study, approximate values of 8KN, 10kN, and 12 kN were selected for different load phases in the simulation.

3.2 DT model development in Simscape

The DT developed in this study is a model-based, high-fidelity simulation framework. It represents a virtual replica of an aircraft EMA, incorporating both electrical and mechanical domains to emulate system-level behaviours under varying operational and fault conditions. The objective of the model is not to replicate a specific certified actuator configuration, but to construct a physics-consistent and diagnostically representative digital twin framework suitable for health management research. The framework is modular, allowing for extensibility and fault injection, and supports the generation of labelled time-series datasets Suitability.

FIGURE 4

FIGURE 5

The electrical architecture follows a 28 Vdc aerospace power bus configuration, which is standard for commercial aircraft actuation systems and consistent with publicly available information from industrial manufacturers (MOOG, 2018). This ensures that the electrical operating envelope reflects realistic aerospace practice. The electromechanical topology, including the motor–gear–leadscrew transmission chain, aligns with validated EMA architectures reported in the literature (Fu et al., 2017; Balaban et al., 2015; Mazzoleni et al., 2021). These prior works provide experimentally and analytically supported modelling frameworks for flight-control EMAs, forming the structural basis of the present digital twin implementation.

The parameter magnitudes listed in Table 1 were selected to remain within the order-of-magnitude ranges reported for aerospace electromechanical actuators. While exact proprietary actuator specifications are not publicly disclosed, the selected voltage level, transmission ratios, and stroke length remain within the typical ranges reported in aerospace EMA literature and industrial documentation (MOOG, 2018; Fu et al., 2017; Balaban et al., 2015; Mazzoleni et al., 2021; Quattrocchi et al., 2021; Annaz and Kaluarachchi, 2023; Wan et al., 2024).

Importantly, behavioural validation is demonstrated through transient electromechanical responses. Under backlash injection, the simulated current surge and corresponding rotor speed dip reproduce the characteristic “current spike–velocity drop” signature observed in experimental PHM testbeds (Balaban et al., 2011). This behavioural consistency provides engineering-level validation of the model’s dynamic response to mechanical disturbances.

The DT architecture is also designed for scalability. Users can adjust fault severity (e.g., magnitude of voltage drop), duration (e.g., transient vs. sustained fault), and onset time (early-stage or end-of-cycle fault) via simulation parameters or script inputs. This capability supports batch generation of scenario-rich datasets with ground-truth labels.

3.3 Fault injection and simulation campaign

where:

FIGURE 6

where:

The interaction occurs naturally through the system’s multi-domain dynamics rather than artificial signal mixing. This preserves physical causality and avoids synthetic feature superposition. The perturbation functions modify model parameters directly within the Simscape multi-domain dynamic equations, allowing the resulting electrical–mechanical interaction to emerge naturally from system coupling.

Crucially, this study moves beyond simple hard faults by simulating a range of fault severities for each mode. For instance, voltage deviations were drawn from a continuous uniform distribution, while backlash and resistance faults were injected at multiple discrete levels to represent varying degrees of degradation.

Each simulation runs for 30 s, covering a full actuator cycle under a predefined load profile. The profile consists of four segments, progressively applying force and displacement as follows: (1) moderate extension, (2) sharp increase in resistance, (3) return under lower load, and (4) final long-travel operation. This structure allows for fault visibility across different loading conditions and speeds.

The timing of fault injection is randomised within controlled intervals. Faults may occur between 5s and 25s of the simulation as shown in Figure 7 to ensure the actuator is under meaningful load. Each run includes metadata: fault type, injection window, magnitude, and expected physical manifestation. For example, in FM3, resistance increases from 0.6Ω to 1.0Ω over a 4-s window, resulting in increased current draw and thermal dissipation.

FIGURE 7

The simulation framework is controlled by MATLAB scripts that automate the entire data generation process. The overall dataset size and construction protocol are detailed in Section 4.3. For each run, the framework generates a time-series data file and a corresponding label file containing the ground truth for the fault type, magnitude, and timing. This automated pipeline ensures the creation of a large-scale, structured, and fully labelled dataset ready for machine learning applications.

4 Results and feature response analysis

The DT model was used to generate multivariate time-series data under each of the seven fault scenarios. The output signals include voltage, current, displacement, velocity, and power. Diagnostic features were extracted from both time and frequency domains to assess their sensitivity and robustness across fault modes.

4.1 Signal trends under individual faults

As illustrated in Figure 8, the injection of a backlash load (black dashed line) into the drivetrain elicits an immediate and sharp response from the system. The motor current (red line) spikes dramatically to over 150 Amps to overcome the sudden mechanical resistance, a stark contrast to the smooth nominal current (light blue). Simultaneously, the angular velocity (dark blue line) experiences a corresponding sharp dip, visualising the classic “current spike, velocity dip” signature that is a symbol of mechanical backlash, jamming or binding events. This provides a clear physical validation that the model accurately captures the electromechanical response to a mechanical fault.

FIGURE 8

4.2 Combined fault responses

Figure 9 demonstrates this by plotting the current response for FM5 under three different levels of resistance. When resistance is high (1.0 Ω, yellow line), the current is generally suppressed and unstable. In contrast, with lower resistance (0.3 Ω, purple line), the current spikes caused by backlash are much sharper and more pronounced. This addresses the issue of fault severity, showing that the diagnostic signature is not static, but changes based on the underlying condition of the components.

FIGURE 9

The FM6 produces the most complex responses, as it combines disturbances from all three domains. Figure 10 deconstructs a single run of an FM6 fault, plotting the individual injected fault components, a voltage drop, a resistance increase, and multiple backlash loads, on the same timeline as the resulting current response. The final current waveform (red line) is a complex superposition of the system’s reactions to each of these stimuli. The large spikes are driven by the backlash events, while the overall shape and magnitude of the current are modulated by the concurrent voltage and resistance faults. This provides a compelling illustration of the diagnostic challenge posed by compound faults, underscoring the need for multi-signal fusion to deconstruct such complex signatures.

FIGURE 10

4.3 Dataset construction and reproducibility

The dataset was constructed through a fully automated simulation campaign to ensure statistical robustness, class balance, and reproducibility. For each of the seven defined operating conditions (Nominal and FM1–FM6), 10,000 independent simulation cycles were executed, resulting in a total of 70,000 labelled samples. Each condition therefore contains an equal number of samples, ensuring strict class balance across all fault categories and preventing bias toward any specific mode during multi-class evaluation.

Simulation and sampling protocol: All simulations were conducted at a fixed sampling frequency of 100 Hz over a 30-s actuator duty cycle, generating 3,000 data points per signal per run. To prevent models from exploiting deterministic phase timing associated with specific load segments, only a single 15-s segment was extracted from each simulation cycle. The starting point of the extracted window was uniformly random within a valid time range, ensuring that at least two operational phases were included. Each sample corresponds to one independently executed simulation run. Only one analysis window is extracted per run, and no overlapping segments are generated. Consequently, no temporal overlap exists between samples in the training and validation subsets, eliminating the risk of data leakage through duplicated time-series fragments. Since fault onset time, magnitude, and duration are randomised independently for each run, substantial intra-class variability is preserved.

Train-validation strategy: For diagnostic evaluation, the dataset was partitioned into 75% training and 25% validation subsets using class-preserving random sampling to maintain equal label distribution in both sets. Because each sample originates from an independent simulation run, the train–validation split does not introduce scenario-level leakage. All classification experiments reported in this study were conducted using this fixed partition protocol to ensure comparability across feature configurations.

Fault magnitude, injection timing, duration, load profile identifier, and random seed values are stored alongside signal outputs. This ensures complete traceability between parameter perturbation and observable electromechanical response.

The modular architecture allows independent adjustment of voltage profiles, mechanical load definitions, and fault severity ranges without modifying the core electromechanical model. When identical random seeds and configuration parameters are used, the dataset can be regenerated deterministically.

Dataset schema: Each simulation run produces one structured CSV file corresponding to a single actuator duty cycle. Signals are recorded in consistent column order and physical units as shown in Table 3.

Feature extraction is performed on the defined 15-s window, as described above.

This structured schema links each time-series sample to its originating perturbation configuration, enabling reproducible benchmarking of diagnostic algorithms and controlled sensitivity studies.

This structured dataset construction and evaluation protocol ensures statistical balance, prevents artificial performance inflation, and provides a reproducible foundation for benchmarking diagnostic algorithms using the generated synthetic EMA dataset.

4.4 Feature extraction and clustering

Figure 11 presents the Principal Component Analysis (PCA) projection of the feature space, which visually confirms the diagnostic separability of the fault modes. The first two principal components capture the majority of the variance, and the distinct, dense clusters for the compound faults (FM5 and FM6) are immediately apparent. FM5–FM6 overlapped partially due to combined signatures but remained distinguishable with higher-order components.

FIGURE 11

UMAP visualisation as shown in Figure 12 further confirmed class separation, suggesting feasibility of supervised fault classification.

FIGURE 12

4.5 Diagnostic performance evaluation

The study considers a fixed, task-specific load profile corresponding to the aerodynamic loading during flap deployment. This design choice reflects a mission-oriented diagnostic setting where the actuator operates under a consistent operational envelope for a given function. Consequently, the present evaluation does not aim to demonstrate cross-task generalisation across fundamentally different load profiles. Instead, practical diagnostic value is assessed via measurement-noise robustness, which directly reflects sensor uncertainty expected in operational monitoring.

To quantitatively assess the diagnostic utility of the DT–generated dataset, supervised classification experiments were conducted using a Random Forest (RF) classifier under additive Gaussian measurement noise ranging from 0% to 5% (0%, 1%, 3%, and 5%) applied to all measured signals. This noise magnitude represents a conservative estimate of sensor uncertainty in operational monitoring environments.

A structured feature-ablation strategy was adopted to examine the extent to which diagnostic information emerges from electromechanical coupling rather than from a single dominant variable.

First, the feature set was limited to three physically meaningful electromechanical variables: mean Current, Rotor Angular Velocity, and Worm Torque Output. These variables directly capture electrical excitation, rotational dynamics, and mechanical load transmission, respectively. The corresponding confusion matrix, shown in Figure 13, exhibits near-perfect classification performance under the same 5% noise condition. Only minor residual confusion remains between closely related compound fault modes. The substantial improvement relative to the single-feature case confirms that diagnostic separability arises from multi-domain physical interactions rather than from isolated signal deviations.

FIGURE 13

To further evaluate engineering-level interpretability, fault modes were grouped into three categories: Nominal, Single Fault (FM1–FM3), and Compound Fault (FM4–FM6). The grouped confusion matrix in Figure 14 demonstrates good separation under the three-feature configuration. This result indicates that, even when individual compound faults may exhibit partial overlap in high-dimensional space, the dataset robustly preserves higher-level diagnostic distinctions relevant to maintenance decision-making.

FIGURE 14

Subsequently, a minimal-sensor scenario was evaluated using only the mean stator current as the input feature. The resulting seven-class confusion matrix is shown in Figure 15.

FIGURE 15

After evaluating the RF classifier, a linear discriminant analysis (LDA) model was also implemented as a simpler linear benchmark. The quantitative accuracy results across varying noise levels are summarised in Table 4. Two key trends are observed. First, the single-feature configuration exhibits limited separability (≈80%), confirming that diagnostic discrimination is non-trivial. Second, the three-feature configuration maintains consistently high accuracy (>96% for LDA and ≈100% for RF) across all tested noise levels, demonstrating robustness against measurement uncertainty. The marginal improvement of RF over LDA indicates that while class boundaries are largely well-separated, non-linear ensemble learning provides additional robustness in resolving compound faults.

This structured analysis provides a clear, data-driven narrative on the diagnostic value of different sensor combinations and validates that the DT-generated data can be used to rigorously evaluate and compare PHM, algorithms.

5 Discussion and implications

The results confirm the validity and utility of the developed DT framework in simulating fault behaviours of EMAs. The observed signal responses align with expected degradation pathways reported in previous literature and failure analysis studies. The framework successfully replicates realistic fault onset, progression, and signal interactions across diverse operational phases.

5.1 Implications for system health management

The presented DT framework offers significant advantages over traditional physical test benches and analytical models. First, it enables exhaustive scenario testing without the cost and risk associated with hardware degradation. Second, it supports fault-specific data synthesis at a scale suitable for machine learning workflows. Finally, it allows the investigation of rarely occurring or mission-critical failure cases under controlled conditions.

Such capabilities are especially valuable for training prognostics models, evaluating health indicators like Remaining Useful Life (RUL), and testing onboard fault-tolerant control strategies. The availability of labelled, high-resolution signal data across fault types strengthens the development of explainable AI for aerospace diagnostics.

5.2 Limitations and model extensibility

While the DT framework demonstrates strong capability in simulating multidomain fault behaviours, several modelling boundaries must be acknowledged.

First, the present model focuses on electromechanical dynamic behaviour under predefined load and voltage profiles. Thermal dynamics, including motor winding heating, gearbox temperature rise, and thermal–mechanical coupling, are not modelled. In practical aerospace systems, temperature variation significantly influences resistance, lubrication properties, and degradation rates. The absence of thermal modelling therefore limits the framework’s ability to simulate temperature-driven fault progression or thermal runaway phenomena.

Second, the implemented fault models represent effect-level perturbations rather than physics-based degradation laws. For example, backlash is introduced as an equivalent reverse torque disturbance, and resistance change is modelled as a discrete parameter shift. While these abstractions preserve physical causality at the system level, they do not capture progressive wear mechanisms such as gradual clearance enlargement, material fatigue accumulation, or corrosion growth over multiple cycles. Incorporating physics-informed degradation evolution laws would enable transition from fault diagnosis toward remaining useful life (RUL) estimation.

Third, parameter uncertainty and manufacturing tolerances are not explicitly considered in the current study. Nominal parameter values were selected within order-of-magnitude ranges reported in aerospace literature; however, no Monte Carlo sensitivity analysis or uncertainty propagation study was conducted. Consequently, the robustness of diagnostic separability under parameter variation remains to be formally quantified.

Finally, the present DT operates as an open-loop simulation framework without real operational data assimilation. In an industrial deployment scenario, the twin could be continuously updated using field measurements through parameter identification techniques, Bayesian updating, or state observers (e.g., Kalman filtering). Such hybrid model–data integration would improve predictive fidelity and enable adaptive recalibration over the actuator lifecycle.

Despite these limitations, the modular structure of the proposed framework allows systematic extension. Thermal sub-models, progressive wear laws, stochastic parameter sampling, and closed-loop control dynamics can be incorporated without restructuring the core architecture. These enhancements will be prioritised in subsequent work to transition the framework from a simulation-driven diagnostic platform toward an operational DT supporting prognostics and lifecycle management.

6 Conclusion

This study presented the development and demonstration of a modular DT framework for simulating faults in aircraft EMAs. By replicating both electrical and mechanical subsystems, the model enables the generation of synthetic datasets for health monitoring research. The fault injection capability covers a broad range of degradation modes, including voltage disturbances, resistance changes, and mechanical backlash—either in isolation or combination.

Through realistic simulation of load profiles and signal responses, the DT captures degradation phenomena such as delayed displacement, increased current draw, and amplitude underperformance. These signatures align with reported failure characteristics in operational aircraft systems. Feature extraction and clustering analysis validated that the simulated data preserves diagnostic fidelity and can support training of classification models.

This DT serves as a flexible, scalable tool for research and development in aerospace prognostics. It bridges the gap between physical test rigs and algorithm design, offering a safe and cost-effective environment to evaluate failure signatures and diagnostic metrics. Future work will integrate thermal, environmental, and closed-loop control elements to further enhance realism and extend the twin’s application to real-time health management and fault-adaptive control strategies.

References

• 1

  AIRBUS (2017). A350 XWB: flight control, flight deck and systems briefing for pilots, 8–9. Available online at: https://www.parlonsaviation.com/wp-content/uploads/2017/12/a350-flight-controls.pdf (Accessed September 7, 2025).

  • Google Scholar

• 2

  AnnazF. Y.KaluarachchiM. M. (2023). Progress in redundant electromechanical actuators for aerospace applications. Aerospace10, 787. 10.3390/aerospace10090787

  • CrossRef
  • Google Scholar

• 3

  ArévaloP.CanoA.JuradoF. (2024). Large-scale integration of renewable energies by 2050 through demand prediction with ANFIS, Ecuador case study. Energy286, 129446. 10.1016/j.energy.2023.129446

  • CrossRef
  • Google Scholar

• 4

  BalabanE.SaxenaA.NarasimhanS.RoychoudhuryI.GoebelK. (2011). Experimental validation of a prognostic health management system for electro-mechanical actuators, infotech@aerospace, 2011. USA: St. Louis, 29–31. 10.2514/6.2011-1518

  • CrossRef
  • Google Scholar

• 5

  BalabanE.SaxenaA.NarasimhanS.RoychoudhuryI.KoopmansM.OttC.et al (2015). Prognostic health-management system development for electromechanical actuators. J. Aerosp. Inf. Syst.12, 2015–2344. 10.2514/1.I010171

  • CrossRef
  • Google Scholar

• 6

  ColeT.TangS. (2022). Liquid metals as soft electromechanical actuators. Mater. Adv.3, 173–185. 10.1039/d1ma00885d

  • CrossRef
  • Google Scholar

• 7

  EmmanouilK. (2020). Reliability in the era of electrification in aviation: a systems approach. Microelectron. Reliab.114 (Nov 2020), 113945. 10.1016/j.microrel.2020.113945

  • CrossRef
  • Google Scholar

• 8

  FuJ.MaréJ.FuY. (2017). Modelling and simulation of flight control electromechanical actuators with special focus on model architecting, multidisciplinary effects and power flows. Chin. J. Aeronautics30 (1), 47–65. 10.1016/j.cja.2016.07.006

  • CrossRef
  • Google Scholar

• 9

  GlaessgenH.StargelD. S. (2012). “The digital twin paradigm for future NASA and U.S. air force vehicles,” in 53rd structure dynamics, and materials conference, 2012.

  • Google Scholar

• 10

  GrievesM. G. (2016). Origins of the digital concept. 10.13140/RG.2.2.26367.61609

  • CrossRef
  • Google Scholar

• 11

  GrievesM. G. (2023). “Digital twins: past, present, and future,” in The digital twin. CrespiN.DrobotA. T.MinervaR. (Cham: Springer). 10.1007/978-3-031-21343-4_4

  • CrossRef
  • Google Scholar

• 12

  JansonR. H.BowmanC.JankovskyA.DysonR.FelderJ. (2017). “Overview of NASA electrified aircraft propulsion research for large subsonic transports,” in 53rd AIAA/SAE/ASEE joint propulsion conference, July 10-12, 2017, Atlanta, Unitesd States. AIAA Propulsion and Energy Forum. 10.2514/6.2017-1701

  • CrossRef
  • Google Scholar

• 13

  LawsonC. P.PointonJ. M. (2008). Thermal management of electromechanical actuator on an all-electric aircraft, 26th international congress of the aeronautical sciences (ICAS), 14-19 September 2008. USA: Anchorage.

  • Google Scholar

• 14

  LiuR.YangB.ZioE.ChenX. (2018). Artificial intelligence for fault diagnosis of rotating machinery: a review. Mech. Syst. Signal Process.108, 33–47. 10.1016/j.ymssp.2018.2.016

  • CrossRef
  • Google Scholar

• 15

  MaréJ.FuJ. (2017). Review on signal-by-wire and power-by-wire actuation for more electric aircraft. Chin. J. Aeronautics30 (3), 857–870. 10.1016/j.cja.2017.03.013

  • CrossRef
  • Google Scholar

• 16

  MazzoleniM.RitoG. D.PrevidiP. (2021). Electro-mechanical actuators for the more electric aircraft. Springer. ISSN, 2193, 1557.

  • Google Scholar

• 17

  MOOG (2018). Modula linear electromechanical actuators. Available online at: https://www.moog.com/content/dam/moog/literature/sdg/space/actuationmechanisms/SAIS_Modular_EMA_Rev_0315.pdf (Accessed: February 19, 2026)

  • Google Scholar

• 18

  MOOG (2020). Primary flight control actuation system for 787. Available online at: https://www.moog.com/products/actuation-systems/aircraft/primary-flight-control-actuation-system-for-787.html (Accessed September 7, 2025).

  • Google Scholar

• 19

  QuattrocchiG.IaconoA.BerriP. C.VedovaM. D. L. D.MaggioreP. (2021). A new method for friction estimation in EMA transmission. Actuators2021, 10–194. 10.3390/act10080194

  • CrossRef
  • Google Scholar

• 20

  WagnerH.NikolovG.BierigA.SpangenbergH. (2011). Challenges for health monitoring of electromechanical flight control actuation systems. SAE Int. J. Aerosp.4 (2), 1316–1323. 10.4271/2011-01-2701

  • CrossRef
  • Google Scholar

• 21

  WanQ.SongC.ZhouY.TongR.MaS.LiuG. (2024). Modelling and analysis of the flap actuation system considering the nonlinear factors of EMA, joint clearance and flexibility. Aerospace11. 10.3390/aerospace1106044

  • CrossRef
  • Google Scholar

• 22

  WangC.FanI. S.KingS. (2022). “Failure mapping for aircraft electrical actuation system health management,” in 7th european conference of the PHM society (PHME 2022), 6-8 July 2022, turin, Italy.

  • Google Scholar

• 23

  WangC.FanI. S.PlastropoulosA. (2025). “Digital twin-based health management for complex aircraft systems: case studies and applications, 2025 IEEE Wireless Communications and Networking Conference (WCNC), March 24-27, 2025, Milan, Italy, IEEE. 1–6. 10.1109/WCNC61545.2025.10978545

  • CrossRef
  • Google Scholar

• 24

  WangC.FanI. S.KingS. (2023). A review of digital twin for vehicle predictive maintenance system, 2023 AeroTech. Fort Worth, USA.

  • Google Scholar

• 25

  XuZ.GaleaM.TigheC.HamitiT.GeradaC.PickeringS. J. (2014). Mechanical and thermal management design of a motor for an aircraft wheel actuator in 17th international conference on electrical machines and systems (ICEMS) (Hangzhou, China), 22–25. 10.1109/ICEMS.2014.7014056

  • CrossRef
  • Google Scholar

• 26

  ZaporozhetsV. I. , K. S.IsaienkoV.SynyloK. (2020). Trends on current and forecasted aircraft hybrid electric architectures and their impact on environment. Energy211 (Nov 2020), 118814. 10.1016/j.energy.2020.118814

  • Pubmed Abstract
  • CrossRef
  • Google Scholar