Table 1 Representative summary of... | Emerald Publishing

Table 1

Representative summary of existing studies on AE-based prognostics and digital twin frameworks

Domain/Focus	Representative studies	Methodology/Contribution	Limitations/Gaps identified
Acoustic Emission (AE) for Structural Health Monitoring (SHM)	Ono (2008), Antonaci et al. (2012), Anastasopoulos et al. (2009), Montalvão et al. (2006), Chou (2024), Shamsudin (2019), Cui et al. (2025)	AE used for crack detection, fatigue, and corrosion monitoring in metals and composites. Experimental and signal-based SHM frameworks demonstrated high sensitivity to early damage	High cost of experimental setups; dependence on sensors and noise-sensitive handcrafted features; limited scalability for real-time deployment
Feature Extraction and AE Signal Processing	Karvelis et al. (2021), Suzuki and Shimamoto (2021), Rummel and Matzkanin (1997), Xi et al. (2018), Li et al. (2015), Liang et al. (2023)	Statistical AE descriptors, FFT/WT analysis, and t-SNE visualization for fault identification	Manual feature design and inconsistent performance across materials; poor temporal learning ability
Digital Twin Frameworks for Predictive Maintenance	Radanliev et al. (2022), Zhong et al. (2023), Kerkeni et al. (2024), Wanasinghe et al. (2020), Xu et al. (2023), Wang et al. (2021), Martinez-Ruedas et al. (2024), Lazakis et al. (2022)	Physics- or simulation-driven twins for equipment condition assessment and lifecycle prediction	Require FEA solvers and calibration; computationally expensive; not suited for adaptive, real-time learning
AI-Driven Prognostics and Deep Learning Models	Kong et al. (2019), Kim and Sohn (2021), Marei and Li (2022), Muneer et al. (2021), Khan et al. (2023), Du et al. (2024), Wang et al. (2020)	CNN/LSTM models for RUL estimation and health classification using vibration or AE data	Trained on limited experimental or FEA data; lack generalization and physical interpretability
Synthetic Data and Physics-Informed Learning	Fabian et al. (2022), Shukla and Deepa (2025), Kim et al. (2022), Ennis and Giurgiutiu (2024), Ciaburro and Iannace (2022)	Generative or physics-informed approaches to produce synthetic datasets for prognostic training	Rarely coupled with digital twins; limited multi-task learning; weak linkage between physics and AI models
Predictive Maintenance Reviews and Industrial Trends	Ponnusamy et al. (2024), Nagy et al. (2025), Garcia et al. (2025), Ucar et al. (2024), Rojas et al. (2025), Shamim et al. (2025)	Comprehensive reviews of AI-enabled predictive maintenance, Industry 4.0, and IoT integration	Identify need for interpretable, data-efficient, and real-time digital twin solutions
Research Gap Addressed by the Present Work	— This Study —	Physics-inspired synthetic AE generation + multi-task CNN–LSTM digital twin + t-SNE/residual interpretability + fully Python-based deployment	Provides unified, data-efficient, and interpretable digital twin framework that removes experimental dependency and achieves real-time scalability

Domain/Focus	Representative studies	Methodology/Contribution	Limitations/Gaps identified
Acoustic Emission (AE) for Structural Health Monitoring (SHM)	Ono (2008), Antonaci et al. (2012), Anastasopoulos et al. (2009), Montalvão et al. (2006), Chou (2024), Shamsudin (2019), Cui et al. (2025)	AE used for crack detection, fatigue, and corrosion monitoring in metals and composites. Experimental and signal-based SHM frameworks demonstrated high sensitivity to early damage	High cost of experimental setups; dependence on sensors and noise-sensitive handcrafted features; limited scalability for real-time deployment
Feature Extraction and AE Signal Processing	Karvelis et al. (2021), Suzuki and Shimamoto (2021), Rummel and Matzkanin (1997), Xi et al. (2018), Li et al. (2015), Liang et al. (2023)	Statistical AE descriptors, FFT/WT analysis, and t-SNE visualization for fault identification	Manual feature design and inconsistent performance across materials; poor temporal learning ability
Digital Twin Frameworks for Predictive Maintenance	Radanliev et al. (2022), Zhong et al. (2023), Kerkeni et al. (2024), Wanasinghe et al. (2020), Xu et al. (2023), Wang et al. (2021), Martinez-Ruedas et al. (2024), Lazakis et al. (2022)	Physics- or simulation-driven twins for equipment condition assessment and lifecycle prediction	Require FEA solvers and calibration; computationally expensive; not suited for adaptive, real-time learning
AI-Driven Prognostics and Deep Learning Models	Kong et al. (2019), Kim and Sohn (2021), Marei and Li (2022), Muneer et al. (2021), Khan et al. (2023), Du et al. (2024), Wang et al. (2020)	CNN/LSTM models for RUL estimation and health classification using vibration or AE data	Trained on limited experimental or FEA data; lack generalization and physical interpretability
Synthetic Data and Physics-Informed Learning	Fabian et al. (2022), Shukla and Deepa (2025), Kim et al. (2022), Ennis and Giurgiutiu (2024), Ciaburro and Iannace (2022)	Generative or physics-informed approaches to produce synthetic datasets for prognostic training	Rarely coupled with digital twins; limited multi-task learning; weak linkage between physics and AI models
Predictive Maintenance Reviews and Industrial Trends	Ponnusamy et al. (2024), Nagy et al. (2025), Garcia et al. (2025), Ucar et al. (2024), Rojas et al. (2025), Shamim et al. (2025)	Comprehensive reviews of AI-enabled predictive maintenance, Industry 4.0, and IoT integration	Identify need for interpretable, data-efficient, and real-time digital twin solutions
Research Gap Addressed by the Present Work	— This Study —	Physics-inspired synthetic AE generation + multi-task CNN–LSTM digital twin + t-SNE/residual interpretability + fully Python-based deployment	Provides unified, data-efficient, and interpretable digital twin framework that removes experimental dependency and achieves real-time scalability