Current Status and Prospects of Digital Twin Approaches in Structural Health Monitoring

Abstract

Structural health monitoring (SHM) is a critical technology for ensuring infrastructure safety, extending their service life, and reducing their maintenance costs. With the rapid development of digital twin (DT) technology, an increasing number of studies have implemented DT in SHM systems. This study provides a detailed analysis of the role of DT in SHM through a comprehensive literature review, specifically examining its applications in damage detection, dynamic response monitoring, and maintenance management. The paper first reviews advances in DT applications across various fields, then systematically discusses how DT enhances monitoring accuracy, enables real-time performance, and supports predictive maintenance strategies in SHM. Finally, technical challenges and future research directions for DT implementation in SHM are explored. The findings highlight DT’s significant potential to improve both the efficiency and the accuracy of structural monitoring systems, while proposing innovative solutions for intelligent infrastructure management.

1. Introduction

Structural health monitoring (SHM), a critical engineering technique, relies on field inspections combined with specialized software and hardware systems for data collection, storage, and processing. SHM plays a vital role in ensuring the safety, reliability, and lifespan of infrastructure systems. However, SHM implementation is a complex task involving interdisciplinary expertise, multiple variables, and dynamic changes. In traditional infrastructure inspections, defect identification, and maintenance strategy development, there is no unified data model to resolve inconsistencies and inefficiencies. These limitations hinder the execution of real-time dynamic analyses, which are increasingly critical for modern infrastructure management [1].

To overcome these challenges, digital twin (DT) technology has emerged as a transformative solution in SHM [2,3,4,5]. Initially conceptualized by Michael Grieves in 2003 and further developed by NASA in 2012, DT fundamentally establishes dynamic maps throughout a physical entity’s lifecycle via multiphysics simulations and real-time data interactions [6]. With technological advancements, the DT framework has evolved to emphasize a “symbiotic connection” between physical assets and their digital counterparts, requiring their continuous synchronization and co-evolution throughout their operational lifespan [7]. This implementation relies on a multi-layered technological architecture: the data layer employs IoT sensors for the real-time acquisition of the physical system states, while the computational layer integrates edge computing for optimized real-time processing and cloud computing for large-scale simulations [8]. The core principle of DT involves establishing a bidirectional mapping between physical and virtual spaces and facilitating advanced simulations, diagnostics, and decision-making processes [9]. As illustrated in Figure 1, the DT framework integrates computational models such as continuous models (CMs), discrete event models (DEMs), statistical models (SMs), and machine learning (ML) techniques to generate comprehensive dynamic representations of physical entities [10]. This integration supports fault mode and effect analysis (FMEA) and prognostics and health management (PHM), both essential for structural resilience and operational efficiency [11].

Furthermore, DT maturity models have been categorized into hierarchical frameworks. A five-dimensional framework encompassing physical entities, data, virtual models, users, and connectivity achieves full-scale health assessments from local components to global structures through finite element analysis and sensor network integration [12]. The three-phase classification of “digital model (DM)–digital shadow (DS)–digital twin (DT)” distinguishes between unidirectional data flows and bidirectional real-time interactions [13]. Emerging concepts like “adaptive DT” and “intelligent DT” leverage machine learning to enable autonomous model optimization, facilitating real-time updates for both physical and digital objects [14].

In infrastructure intelligence, DT integration with IoT and augmented reality (AR) demonstrates significant potential. For instance, the DT model of Milan Cathedral enables the real-time analysis of complex structural behaviors, evolutionary patterns, and future performance predictions [15]. In smart city applications, DT implementations incorporate geospatial mapping, BIM, 5G-enabled IoT, blockchain, and collaborative computing, with demonstrated use cases in intelligent healthcare and transportation systems via blockchain integration [16]. Case studies such as those on HVAC anomaly detection systems further validate DT’s practical utility in building asset management [17].

Practically, DT applications have expanded from aerospace and manufacturing to civil engineering and urban systems. In aerospace, DT-based life prediction models for aircraft structures utilize real-time stress distribution monitoring and fatigue accumulation analysis, significantly reducing the operational and maintenance costs [18]. Manufacturing applications achieve standardized data formats and optimized information flows, as exemplified by industrial cutting tool monitoring systems [19]. Breakthroughs in civil engineering include post-earthquake structural assessment frameworks that correlate BIM-optimized damage detection with specific building components [20] and bridge maintenance systems integrating 3D digital models with real-time sensor data for prestressed concrete structures [21].

In conclusion, DT technology establishes a tight coupling between the physical and the digital domains, enabling intelligent monitoring, prediction, and decision-making. This positions DT as a critical enabler for next-generation smart infrastructure development.

This study explores the application of DT technology in SHM through three interconnected themes. DT enhances real-time damage detection and identification by synchronizing sensor data with virtual models, enabling an early and accurate diagnosis. Additionally, DT plays a key role in dynamic response monitoring and analysis by simulating a structure behavior under varying loads and environmental conditions, thereby improving the accuracy of performance predictions [22,23]. Furthermore, DT supports maintenance management and decision-making through the integration of historical and real-time data, optimizing predictive maintenance strategies and reducing the downtime [24,25]. Collectively, these applications demonstrate DT’s capability to bridge physical and virtual systems in SHM, facilitating dynamic monitoring, simulation, and data-driven decision-making.

This research focuses on the application of DT technology in SHM, aiming to provide a comprehensive review of its capabilities and potential to address the limitations of traditional SHM methods (Figure 2). Specifically, it investigates how DT enhances real-time monitoring, predictive maintenance, and decision-making processes in SHM. By leveraging DT technology, this study aims to overcome key challenges in structural damage detection, dynamic response analysis, and maintenance management, thereby providing directions for future research in the DT technology for SHM.

2. Damage Detection and Identification

Structural damage detection (SDD) constitutes a pivotal research domain within SHM. Over the service life of structures, exposure to adverse environmental conditions, operational loads, and other external factors significantly increases the risks of aging, damage, and even sudden collapse [26]. The existing SHM technologies have achieved substantial advancements in surface defect detection, particularly through AI-based automated techniques encompassing defect classification (e.g., the use of convolutional neural networks (CNNs) for automated defect classification on concrete bridge surfaces, enabling high-precision differentiation between crack types, spalling types, and corrosion types [27]), localization (e.g., the application of cycle-consistent generative adversarial networks (CycleGANs) for data transformation from undamaged to damaged states, providing accurate crack localization within SHM, which is crucial for targeted maintenance and repair [25]), and pixel-level segmentation (e.g., the use of fully convolutional networks (FCNs) for the pixel-level segmentation of concrete structures, allowing for detailed crack identification and quantification with submillimeter precision [28]). These studies demonstrate that AI technologies significantly enhance the efficiency and accuracy of defect detection and analysis in SHM tasks. However, detecting internal structural damage remains a significant challenge.

In complex operational environments, vibration-based damage detection techniques are widely adopted in structural systems. For instance, damage identification in bridge structures under temperature variation interference has been achieved through the integration of time series analysis and artificial neural networks [29]. The effectiveness of distributed SDD based on two-dimensional (2D) convolutional neural networks (CNNs) has been validated for steel frame structures [30]. In addition, the tap–scan method, which analyzes the spectral characteristics of tapping signals, has been employed to identify damage location and severity in continuous beam structures [31]. These approaches primarily analyze alterations in structural vibration characteristics—such as natural frequencies, mode shapes, and modal damping—derived from vibration signals [32]. Nevertheless, their efficacy is constrained by environmental variability, measurement inaccuracies, and inefficiencies in the processing of large-scale monitoring datasets [33]. In contrast, DT technology creates a real-time digital representation of a structure, facilitating precise damage detection and overcoming the limitations of traditional vibration analysis [34].

To bridge the gap between traditional methods and emerging digital solutions, various damage detection methods with distinct advantages and limitations have been developed [35]. These methods are increasingly integrated within the DT framework to enable holistic, real-time structural health assessments. This study compares these methods, highlights their strengths and weaknesses, and explores DT integration strategies to address challenges. Each method can be incorporated into a DT framework to enhance SHM effectiveness, as they complement each other in addressing specific damage detection aspects. DT has emerged as a promising solution [36], enabling real-time evaluations and precise damage identification through lifecycle-accurate digital models [37]. However, real-time computational demands—particularly when processing large-scale sensor data—remain a significant implementation challenge [38].

A summary of DT applications in structural damage detection and identification is presented in Figure 3. Within DT technology, SDD evolves through six interconnected approaches (Table 1), unified under DT to integrate physical data, simulation models, and real-time monitoring into a dynamic detection system. With a comprehensive SHM review, we identify three key synergies: (1) data-driven methods mitigate the computational constraints of finite element methods (FEMs); (2) unmanned aerial vehicle (UAV) and building information modeling (BIM) fusion enables multiscale visualization; (3) hybrid Bayesian frameworks bridge simulation–reality gaps. This taxonomy highlights the methods’ complementarity and their integration within DT frameworks to address complex damage detection problems.

2.1. Data-Driven Approaches

In recent years, data-driven simulation models have gained extensive application in structural damage detection due to their adaptability and efficiency [52]. These models provide innovative solutions by extracting features from real or simulated data to facilitate damage identification [53]. However, acquiring real-world damage data in industrial environments is often difficult or even impractical [54]. To address this limitation, artificially generated damage data have emerged as an effective substitute [55]. The integration of data-driven approaches within the DT framework—utilizing simulated data and machine learning (ML) models—compensates for the scarcity of real-world damage data and enhances predictive accuracy [56]. Furthermore, DT integrated with ML models generates simulated data, effectively alleviating the shortage of real damage data [57].

Data-driven methods leveraging DT technology significantly improve the accuracy and applicability of damage detection [58]. For instance, a DT model for cable-stayed bridges was developed using a nonlinear model updating approach to assess collapse vulnerability [59]. This method demonstrates superior predictive capabilities compared to traditional vibration-based techniques, which are limited by measurement inaccuracies and environmental variability. In another study, a neural network-based DT model integrates fiber optic sensors (FOSs) with deep neural networks to convert sensor data into interactive visualizations. Validation via finite element analysis (FEA) confirmed the model’s accuracy in capturing structural data, with an average bending moment error below 2 kN·m and a deflection absolute error under 0.15 mm. Furthermore, the integrated artificial neural network (ANN) exhibited a mean absolute strain error of less than 1 με during testing, validating the precision of the DT-generated baseline strain data [60]. This exemplifies how data-driven DT techniques enable real-time sensor data processing for enhanced damage detection.

Moreover, a system integrating physical models with ML techniques has been developed for structural damage detection. In this system, physical models provide the theoretical foundation for structural behavior, while ML algorithms learn from extensive historical data to identify damage patterns [61]. A novel deep learning approach was developed for prestressed concrete (PC) beam bridge damage assessment, addressing data acquisition and adaptability challenges. This method employs complete ensemble empirical mode decomposition (CEEMD) and frequency domain decomposition (FDD) to process vibration response data, followed by a deep convolution long short-term memory neural network (DCLSTMNN) for damage detection. The approach demonstrated superior accuracy in identifying damage location and severity under uncertain conditions [39]. Despite its promise, the computational burden of training deep learning models on large datasets remains a limitation, and real-time processing with retained accuracy poses ongoing challenges. Another study developed DT models for two UK railway bridges using data-driven and physics-based model-driven methods. The merged system employed a data-driven engineering approach, significantly enhancing DT accuracy and applicability in SHM [40].

Automated image analysis technologies have advanced structural damage detection. For example, robotic systems minimize the risks associated with manual inspections, while captured data are analyzed to generate corrosion, delamination, and elastic modulus maps [62]. Although promising, these systems face challenges in real-time feedback and efficient large-scale datasets processing, particularly for infrastructure. An automated robotic bridge inspection platform has also been proposed, reducing operator intervention [63]. Advancements in automated data collection further enable the systematic optimization of bridge management system (BMS) databases [64].

The data-driven methods focus on extracting meaningful information from sensor data to assess structural health. While digital models can generate sufficient data for various health states, domain shift issues between simulated (source domain) and real-world (target domain) data persist [65,66]. To mitigate this, transfer learning (TL) techniques adapt deep learning models to the target domain. A novel structural damage detection method combines digital models with TL, using neural networks to learn from simulated monitoring data [26]. Furthermore, physics-informed transfer learning effectively addressed domain discrepancies by constructing empirical physical models from data, enhancing model interpretability [67]. However, TL faces challenges in real-time domain adaptation, especially with large-scale sensor data.

Lastly, a feature transfer learning method for truss bridge damage identification has improved the traditional ML model generalization across bridge structures. By extracting shared features from simulated and real-world data, this approach enables knowledge transfer [68]. To address the issue of insufficient training data and limited generalization in impact detection, a transfer autoencoder (transfer-AE) network combining generative TL and digital models was developed. Validated on a dome structure, transfer-AE eliminates systematic errors between numerical and physical systems through domain-invariant feature learning, achieving an F1 score of 0.85 and a mean absolute error (MAE) of 0.95 [41]. While promising, its real-time computational complexity hinders its deployment in large infrastructure monitoring systems.

2.2. BIM-Based Approaches

Building information modeling (BIM) serves as a digital representation of structures’ physical and operational attributes. With robust information modeling and visualization capabilities, BIM has become a cornerstone for smart structural management and SHM systems [69,70,71]. When integrated with DT, BIM enables dynamic structural model updates based on real-time sensor data, supporting continuous monitoring and decision-making [72]. However, heterogeneous data integration may cause delays or inaccuracies in model updates [73].

BIM-SHM integration aims to embed monitoring data into BIM models, updating attributes to evaluate structural performance [74]. Current BIM-based SHM systems predominantly rely on offline processing to analyze heterogeneous sensor data (e.g., vibration sensors) and visualize structural behavior through BIM software. Real-time BIM model updates within the DT framework introduce computational complexity, often requiring trade-offs between accuracy and efficiency [75]. Despite these challenges, such systems hold significant potential. For instance, a bridge risk assessment model integrated into a BIM platform created a digital twin, enhancing extreme weather risk management [42]. A machine learning-based method for the automatic semantic segmentation of bridge components also improved detection accuracy [43].

2.3. Bayesian Model Updating and Augmented Reality (AR) Applications

Bayesian model updating reduces discrepancies between measured and predicted responses [39]. This technique integrates sensor data into DT frameworks to dynamically adjust virtual models based on the actual structural behavior, improving damage detection accuracy [76]. Integrating data from the physical twin into the model optimizes decision-making in SHM, maintenance, and management, thereby improving the model accuracy. A framework combining physical models with ML employs dynamic Bayesian networks to assimilate vibration data and extract damage-sensitive features for real-time structural health assessment [44]. Another study introduced a continuous Bayesian updating technique integrating both linear and nonlinear finite element models (FEMs) with soil–structure interaction effects and ground motions, improving the virtual model alignment with real structures [77]. However, damage visualization remains challenging, and FEM updates relying on simplified assumptions (e.g., linearity) may yield inaccurate simulation results.

Augmented reality (AR) technology is increasingly applied in SHM. An AR-based system integrating servers, sensors, and AR headsets enable real-time data transmission and visualization, assisting inspectors [78]. A feature-based reverse propagation refinement scheme (f-BRS) combined with HoloLens 2 (HL2) headsets calculates damaged area sizes. The system isolates damaged regions via gesture inputs, transmits three-dimensional (3D) coordinates and video to a server, and applies f-BRS segmentation with deep neural networks (DNNs) to generate binary masks and quantify damage areas [79].

2.4. Finite Element-Based Approaches

The finite element method (FEM) is a primary numerical technique for implementing DT technology in SHM [45]. Virtual twin parameters are calibrated based on data from a physical structure. Following model validation, the virtual twin (an FEM model) generates large datasets that interact bidirectionally with the physical asset measurement data [80]. FEM and DT exhibit strong complementarity: FEM provides detailed structural behavior simulations, while DT updates models with real-time sensor data to enhance the predictive capabilities [76]. The application framework is illustrated in Figure 4. Consequently, FEM enables precise structural responses’ predictions and supports damage detection [26]. However, FEM’s significant computational demands are a challenge for real-time monitoring applications, as DT-driven FEM updates require substantial processing resources.

A data-driven FEM approach has been proposed to predict structural damage states and quantify uncertainties using updated FEM models. Labeled damage data were generated for a hinged composite truss structure, with model-based approaches substituting vibration measurements to address data acquisition challenges [81]. FEM-based DT modeling was applied to a cable-stayed bridge scale model for collapse vulnerability assessment [46]. The inverse FEM (iFEM) method—which reconstructs displacement and stress fields from strain measurements—was combined with effect superposition techniques to optimize prediction accuracy, providing a robust SHM framework [47].

Despite FEM’s strong damage detection performance, its computational complexity limits real-time monitoring. To enhance DT interactivity, researchers are investigating distinctions between DT and numerical models. Customized digital simulations capture potential damage characteristics, surpassing traditional offline models. FEM-based DT has demonstrated damage detection ability in scaled truss bridges [82], while a fatigue assessment method integrates monitored and unmonitored data for orthotropic steel bridge decks [9].

Advancements include marker cluster-based DT for plate and girder-plate bridge damage detection [83,84]. A self-anchored suspension bridge DT constructed with SHM data showed a higher vulnerability curve accuracy than design-based finite element (FE) models [85].

2.5. UAV-Based Methods

With rapid advancements in unmanned aerial vehicle (UAV) technology, UAV-based SHM has emerged as a transformative approach [4]. UAVs capture high-resolution images processed into 3D digital point clouds via photogrammetric techniques, enabling virtual twin creation and textured 3D scene generation [86]. Integrated with the DT framework, UAVs enhance real-time visual monitoring through scalable structural data [87]. As a result, UAVs have become a valuable tool in infrastructure monitoring, offering an efficient, non-invasive solution for assessing structural integrity. For instance, a post-earthquake evaluation framework combines UAV imagery, component recognition, and damage assessment [20]. In addition, a UAV–bridge information modeling (BrIM) system automates concrete bridge deterioration detection using computer vision, as shown in Figure 5, comprising three phases: 3D model development, UAS data processing for defect detection, and cloud-based model integration for inspections [88]. However, processing UAV-collected high-resolution imagery requires significant computational resources, causing model update delays [89].

Cracks are a critical indicator of structural health, as their presence can lead to reinforcement corrosion, significantly impacting a structure durability and stability [27]. In this context, UAV-based methods increasingly improve crack detection accuracy [87]. The crack boundary refinement framework (CBRF) employs multi-scale cascading and the active boundary loss function, outperforming traditional methods. High-resolution imaging enables a safe structural surface inspection at 1.5 m, significantly reducing the inspection time [48]. UAVs also facilitate automatic crack segmentation, highlighting big data’s role in structural assessment [27]. Continuous UAV advancements support real-time large-scale monitoring, providing deeper structural insights [87]. Permanent 3D geometric records are vital for long-term monitoring [90]. An integrated crack detection method using virtual reality (VR) and 3D laser scanning automatically identifies concrete cracks, enhanced by VR visualization [91]. The high-resolution crack segmentation framework (HRCSF) balances memory and computational load through multi-scale feature extraction, achieving superior performance [92].

Automated UAV detection with high-resolution imaging is becoming a dominant trend in bridge monitoring [87]. To improve detection accuracy, some studies have incorporated laser ranging devices for enhanced data collection [49,93,94,95]. For instance, one study mounted lasers on UAVs to measure object distances and pixel resolution during inspections [49]. Furthermore, post-disaster image techniques enable rapid reinforced concrete bridge evaluations [93]. Another study advanced this approach by integrating cameras with laser rangefinders into a multimodal data acquisition system for crack detection, demonstrating the ability to measure cracks wider than 0.1 mm and achieving 92% accuracy in crack length quantification [94]. Notably, the ground-penetrating radar (GPR) has been successfully integrated with UAVs to provide detailed assessments of bridge deck conditions, significantly improving the accuracy of defect detection and localization [95].

2.6. 3D Model Reconstruction Using Laser Scanning

Light detection and ranging (LiDAR) and UAV photogrammetry are key technologies for high-precision digital modeling [96]. Their point clouds enable 3D structural reconstructions critical for damage assessment. However, their real-time point cloud integration into DT models faces computational bottlenecks, as high-density data require balancing accuracy and processing speed. LiDAR was first integrated with VR for reinforced concrete inspections, capturing defect-rich 3D surface images for immersive analysis [50]. Heuristic algorithms automate the semantic segmentation of steel beam bridge laser point clouds, further advancing the use of laser scanning in bridge inspection and management [51]. The need for the real-time processing of LiDAR data in conjunction with other sensor data often results in delays or inaccuracies in model updates, affecting the precision of the digital twin model.

A comprehensive approach for the quality analysis of digital point clouds generated from various techniques, such as image acquisition and laser scanning, has been introduced. The resulting 3D models, which feature a high point density, are considered digital twin models. However, challenges remain in integrating real-time monitoring data, and knowledge mining from 3D surface models has certain limitations [97]. As a solution, a method for digitally fitting bridge objects to represent their true geometric shape, as well as automatically converting point cloud data into 3D models, has been proposed [83].

This section provides an overview of the applications and development trends of digital twin technology in structural damage detection and identification. From data-driven simulation methods to applications based on BIM, finite element models, and UAV technology, digital twin technology plays an increasingly important role in enhancing the accuracy, efficiency, and real-time capabilities of structural health monitoring. Despite challenges related to data acquisition and computational complexity, the potential of digital twins in structural health monitoring has been progressively validated, with expectations for broader applications in the future. The comparative analysis in Table 1 highlights the strengths and limitations of each method, emphasizing their complementary roles within the DT framework. This integration enables a more comprehensive and dynamic approach to SHM, addressing both immediate and long-term challenges in infrastructure management.

Table 1. Comparative analysis of digital twin-enabled structural damage detection methods.

Dimension	Optimal Method	Strengths	Weaknesses
Real-time performance	Data-driven methods (Section 2.1)	Lightweight machine learning models (e.g., DCLSTMNN [39]) enable edge computing	High dependency on data quality, and domain shift issues between simulated and real-world data
Scalability	BIM-based approaches (Section 2.2)	Standardized BIM supports multi-project integration (e.g., bridge group [42])	Real-time model update increases complexity; heterogeneous data delays
Adaptability to complex scenarios	Bayesian model updating and AR applications (Section 2.3)	Dynamically integrates multi-source data (e.g., vibration response and visual data [44])	High computational resource demands for AR visualization; limited generalizability across structures
High-precision simulation capability	Finite element method (FEM) (Section 2.4)	Complex boundary conditions and nonlinear analysis (e.g., collapse vulnerability assessment of cable-stayed bridges [46]) and provides response prediction [47]	High computational complexity limits real-time application; requires validated physical parameters
Cost-effectiveness	UAV-based methods (Section 2.5)	High-resolution imaging (e.g., collect UAV data and apply computer vision algorithms to process UAV images for defect detection [88])	Computational bottlenecks in processing high-resolution images, limited accuracy in dynamic load scenarios
Accuracy	3D laser scanning (Section 2.6)	LiDAR technology provides sub-millimeter geometric accuracy (e.g., concrete surface defect detection [50]), suitable for static structural detail analysis [95]	Real-time integration of dense point clouds is computationally intensive; limited to surface defect detection

Table 1. Comparative analysis of digital twin-enabled structural damage detection methods.

Dimension	Optimal Method	Strengths	Weaknesses
Real-time performance	Data-driven methods (Section 2.1)	Lightweight machine learning models (e.g., DCLSTMNN [39]) enable edge computing	High dependency on data quality, and domain shift issues between simulated and real-world data
Scalability	BIM-based approaches (Section 2.2)	Standardized BIM supports multi-project integration (e.g., bridge group [42])	Real-time model update increases complexity; heterogeneous data delays
Adaptability to complex scenarios	Bayesian model updating and AR applications (Section 2.3)	Dynamically integrates multi-source data (e.g., vibration response and visual data [44])	High computational resource demands for AR visualization; limited generalizability across structures
High-precision simulation capability	Finite element method (FEM) (Section 2.4)	Complex boundary conditions and nonlinear analysis (e.g., collapse vulnerability assessment of cable-stayed bridges [46]) and provides response prediction [47]	High computational complexity limits real-time application; requires validated physical parameters
Cost-effectiveness	UAV-based methods (Section 2.5)	High-resolution imaging (e.g., collect UAV data and apply computer vision algorithms to process UAV images for defect detection [88])	Computational bottlenecks in processing high-resolution images, limited accuracy in dynamic load scenarios
Accuracy	3D laser scanning (Section 2.6)	LiDAR technology provides sub-millimeter geometric accuracy (e.g., concrete surface defect detection [50]), suitable for static structural detail analysis [95]	Real-time integration of dense point clouds is computationally intensive; limited to surface defect detection

3. Dynamic Response Monitoring and Analysis

Traditional structural health monitoring methods rely on contact sensors to collect structural response data, which are compared against safety thresholds to assess damage states [98]. However, these methods face three critical limitations:

(1)

Sensor placement issues: sensors are typically placed empirically in critical areas, limiting the monitoring of unsensed components.

(2)

Environmental and human influences: contact sensors are vulnerable to environmental factors (e.g., wind, rain, snow, and electromagnetic interference) and human-related issues, leading to gradual declines in accuracy and reliability over time.

(3)

Lack of fault prediction: contact sensors lack predictive capabilities, making it difficult to estimate the remaining useful life of a structure.

To address these limitations, DT technology has been integrated into SHM, offering significant improvements over traditional methods. Unlike static BIM, which represents only geometric data, DT creates dynamic response models that simulate real-time structural behavior. These models integrate multidisciplinary, multi-physics, and multi-scale simulation processes by combining physical models, sensor data, operational history, and other parameters [99]. DT’s real-time simulation capabilities enable continuous state reflection and dynamic optimization, overcoming sensor placement issues limitations.

3.1. Dynamic Response Monitoring and Applications of Digital Twin

DT technology has demonstrated significant potential in dynamic response monitoring [32]. A finite element analysis (FEA)-based computational engine was developed to address the high computational costs and poor real-time performance of traditional physics-based simulations for complex, large-scale structures. By integrating high-performance computing and optimization algorithms, this engine enables the efficient and accurate real-time simulation and monitoring of large structures, such as bridges, and their continuous assessment. This is critical to overcome the limitations of traditional methods that cannot efficiently update the model parameters in real-time based on live sensor data. However, balancing real-time performance with simulation accuracy remains challenging due to computational burdens in large-scale applications. Studies have shown that this DT-based computational engine significantly reduces the computational burden and enhances the monitoring performance in large structures [100].

Recent studies have advanced the application of DT systems in real-time structural response monitoring for port infrastructure. One such development involves UAV-based systems for port vulnerability assessment, which efficiently collect detailed images and data from hard-to-reach areas. The 3D models generated by these UAVs enable more accurate structural health assessments and the identification of potential vulnerabilities [101]. Another system integrates artificial neural networks (ANNs) with sensor data to predict and analyze the structural behavior of port structures, effectively capturing structural performance changes under complex environmental conditions, thereby enhancing the real-time monitoring capabilities [102]. Furthermore, research has explored the use of a Gaussian mixture model (GMM)-based adaptive principal component analysis (PCA) algorithm for monitoring parameters such as slope, settlement, and spacing. By combining multiple sensors, this system is capable of detecting anomalies in real time, even under varying environmental conditions [103].

Additionally, a DT system integrating FEA with machine learning techniques has been developed to continuously incorporate sensor data and achieve real-time structural condition prediction for bridges, dams, levees, and other similar structures using a trained deep neural network model [104]. This integration further strengthens DT’s ability to process large volumes of sensor data and make real-time predictions, significantly improving damage detection and condition forecasting. The challenge of processing large datasets while maintaining real-time performance remains, as larger datasets require more computational resources and time to process, thus presenting significant scalability issues when transitioning from small-scale to large-scale infrastructure.

3.2. Integration of DT with Other Advanced Technologies

The integration of DT with other advanced technologies is advancing SHM, offering smarter and more efficient solutions. This integration represents a key shift in the evolution of SHM, enabling more accurate assessments and predictive maintenance potential across infrastructure systems [105]. For instance, a method integrating bridge information modeling (BrIM) derived from terrestrial laser scanning (TLS) with DT has been proposed to address challenges in data acquisition and model updates in traditional bridge monitoring and maintenance. TLS captures high-precision geometric data, which are integrated into SHM systems for real-time structural condition assessments. This integration improves bridge assessment accuracy and facilitates predictive maintenance by ensuring real-time data constantly integrated into the monitoring process, enhancing the DT model’s accuracy [106].

In a similar vein, a spatial structure DT approach applied to smart building and facility management addresses challenges such as data fragmentation, integration difficulties, and the absence of real-time monitoring in conventional practices. This approach has become particularly relevant as the demand for smarter building systems and real-time monitoring solutions continues to grow. By constructing a high-precision 3D digital model of physical spaces and integrating it with an internet of things (IoT) sensor network, this method enables real-time monitoring and precise management throughout a building’s lifecycle [98]. The integration of IoT with DT enhances a building’s ability to adapt to real-time changes, significantly improving monitoring accuracy and operational efficiency. However, issues such as data synchronization, fault tolerance, and high computational cost must be managed for large-scale deployments.

Further advancements in SHM include the integration of modeling and data fusion within a DT framework. These advancements highlight the ongoing push for more holistic and automated SHM systems that combine physical modeling with real-time data to improve decision-making. This approach combines physical model simulations with real-time sensor data and employs deep learning algorithms to improve the accuracy and timeliness of structural health monitoring, enabling a more predictive and automated monitoring process [107]. For instance, the combination of augmented reality (AR) technology with advanced nondestructive testing (NDT) methods, such as ground-penetrating radars (GPRs), laser distance sensors (LDSs), infrared thermography (IRT), and thermographic cameras (TCs), provides a comprehensive approach for real-time monitoring and management [108]. Furthermore, the DF-CDM model, which integrates finite element method (FEM) simulations with real-time sensor data, significantly improves the structural dynamic response reconstruction accuracy, reducing prediction errors by 42% while enhancing computational efficiency [109]. Collectively, these technological developments have led to substantial advancements in SHM systems, providing more accurate, timely, and reliable structural health assessments.

Additionally, a novel DT modeling technique employing the statistical finite element method (SFEM) has been introduced to model self-sensing structures. This method addresses traditional FEA limitations in predicting structural behavior under uncertainty. By incorporating statistical analysis with sensor data, this technique enhances predictive accuracy in scenarios where traditional methods struggle. By integrating sensor data with statistical analysis, SFEM yields more accurate behavior representations and enables real-time damage detection, overcoming challenges in uncertain environments [110]. However, applying SFEM within a DT framework requires complex data processing, which may delay real-time monitoring and predictive assessments, highlighting the trade-off between accuracy and computational efficiency.

Together, these studies demonstrate that integrating DT technology with other advanced methods significantly enhances real-time performance, accuracy, and data integration in traditional SHM systems, offering more precise and flexible solutions. These integrated solutions are poised to revolutionize SHM, making systems more intelligent, automated, and predictive.

Table 2. Applications of digital twin in dynamic response monitoring.

References	Methodologies and Contributions
Wang et al. [98]	DT of spatial structures in smart buildings to address data fragmentation and integration
Zhang et al. [100]	FEA-based computational engine for real-time simulation of large, complex structures
Tsaimou et al. [101]	Designed a UAV-based system for port vulnerability assessment, generating 3D models from UAV imagery to identify structural weaknesses in inaccessible areas
Jayasinghe et al. [102]	Integration of ANN for high-precision real-time response monitoring of port structures
Lee et al. [103]	Proposed a GMM-PCA algorithm for multi-sensor anomaly detection in slope and settlement monitoring, enabling real-time alerts under varying environmental conditions
Michael et al. [104]	FEA and ML integration for real-time sensor data fusion and structural state prediction using DNN
Mohammadi et al. [106]	TLS-BrIM integration with DT for high-precision bridge structure evaluation
Sun et al. [107]	DT-based SHM integrating physical model simulations with real-time sensor data
Kilic et al. [108]	Combined AR with advanced NDT methods (GPR, IRT) for comprehensive real-time monitoring, improving defect visualization and decision-making
Shu et al. [109]	Introduced the DF-CDM model (FEM + sensor data) to reconstruct structural dynamic responses, reducing prediction errors by 42% while enhancing computational efficiency
Febrianto et al. [110]	Self-sensing structural modeling using SFEM to enhance prediction accuracy

summarizes the studies discussed above, while

Table 3. Comparative analysis of DT approaches for SHM.

Approach	Strengths	Weaknesses	Suitable Applications
IoT-integrated DT [98]	High accuracy in data integration supports lifecycle management	High computational cost; limited real-time performance due to data sync delays	Large bridges, high-rise buildings
FEA-based DT [100]	Captures complex structural behavior; validated for large-scale models	Requires extensive datasets	Port structures, complex environments
UAV-DT fusion [101]	Efficient data collection in hard-to-reach areas; cost-effective	Limited to surface defects; image-processing delays in real-time	Post-disaster inspection, coastal infrastructure
ANN-based DT [102]	Handles environmental uncertainties; high prediction accuracy	Sensitive to data quality; computationally intensive	Self-sensing structures, uncertain environments
GMM-PCA DT [103]	Robust against environmental variability; real-time anomaly detection	Limited to static parameter monitoring; requires sensor calibration	Slope stability, settlement monitoring
FEA-ML-integrated DT [104]	Combines physics-based and data-driven insights	Requires extensive training data	Complex dynamic structures, real-time state prediction
TLS-BrIM-integrated DT [106]	High geometric precision; supports predictive maintenance	Dependent on sensor network reliability; data security risks	Smart buildings, facility management
Modeling data–DT fusion [107]	Predictive automation; holistic monitoring	High computational demand; complex integration of heterogeneous data	Large infrastructures requiring holistic monitoring (e.g., dams, power plants)
AR-NDT DT [108]	Enhanced visualization of defects	High hardware costs (e.g., AR headsets); limited field applicability	Critical component inspection
DF-CDM DT [109]	Reduces prediction errors by 42% and improves computational efficiency	Requires validated FEM parameters; limited to linear dynamic systems	Bridges under dynamic loads
SFEM-based DT [110]	High-precision uncertainty modeling	Expensive equipment and complex data processing pipelines	Bridge monitoring and maintenance

presents a comparative analysis of the pros and cons of each approach, highlighting their suitability for different SHM applications.

4. Maintenance Management and Decision Support

Structural deterioration has become an increasingly significant global challenge, with diverse maintenance and management issues emerging [111]. Traditional methods, which rely heavily on documentation and visual inspections, fail to detect latent damage until it progresses into major problems [78]. Moreover, these methods often face computational complexity and inefficiency when addressing material nonlinearity. To address these challenges, researchers have developed innovative approaches that enhance both accuracy and efficiency by simplifying material nonlinearity in double nonlinear analysis for calculating the ultimate bearing capacity of in-service reinforced-concrete (RC) arch bridges. These methods achieve significant computational efficiency improvements while maintaining accuracy comparable to that of traditional approaches [112]. Figure 6 compares the workflows of traditional maintenance, SHM, and key performance indicator (KPI)-based DT maintenance. Documentation-based inspections are additionally prone to data loss, time-consuming analyses, and operational interruptions, resulting in reactive maintenance, which often accelerates deterioration and service disruptions.

4.1. Application of Digital Twin in Maintenance Management

With advancements in modern information technologies, DT has become essential for SHM. High-precision DT models enable the continuous exchange and updating of real-time monitoring data between physical structures and their digital counterparts, facilitating comprehensive data feedback and dynamic analysis. Specifically, DT models are constructed using parameters such as sensor data, mechanical performance metrics (e.g., stress, strain, temperature), environmental factors (e.g., temperature fluctuations, humidity), material properties (e.g., tensile strength, elasticity), and operational history (e.g., usage cycles, maintenance records). This integration allows stakeholders to gain real-time insights into the structural conditions, optimizing the maintenance decisions [21]. For example, structural stress variations, fatigue accumulation, and environmental loading can be predicted through historical and real-time data analysis, enabling a proactive maintenance scheduling to extend a structure’s service life and reduce costs [10]. However, integrating large datasets and continuous model updates poses real-time computational challenges, demanding a high processing power. Accurate models often require intensive computations, impacting the real-time performance for large-scale structures [113]. Balancing model complexity with real-time computational constraints remains a core challenge in the widespread application of DT for maintenance management, which further exacerbates the scalability of such systems in real-world applications.

A real-time SHM system that integrates internet of things (IoT), BIM, FEA, and field loading tests has been proposed to optimize the monitoring process. However, ambiguities in IoT data collection and processing limit the reliability of BIM-based SHM systems [113]. To address this, a BIM-based DT system integrates intelligent sensors and wireless communication technologies for high-speed data transmission, processing, and visualization, enabling early structural health warnings through predictive analytics [114].

Moreover, a DT model for bridges integrates structural stress characteristics, material fatigue, and environmental load predictions, incorporating parameters such as material aging and usage history for real-time feedback [111]. Additionally, a digital twin-enhanced intelligent maintenance (DTIM) framework addresses corrosion–fatigue deterioration in bridge hangers using monitoring data and adaptive algorithms, enabling customized repairs based on real-time degradation assessments [115].

4.2. Interdisciplinary Applications of Digital Twin Technology

Numerous studies have successfully integrated models and data from multiple disciplines. This interdisciplinary integration is transforming SHM, enabling more comprehensive, efficient, and precise approaches to maintenance management across various types of infrastructure [116]. For instance, a DT-based bridge maintenance system combines 3D information models, digital inspection systems, and image processing for prestressed concrete bridges. This system integrates parameters such as physical dimensions, material properties, and vibration modes, continuously updated via DT for early damage detection [21]. Additionally, a deep learning-based framework optimizes 3D modeling and DT integration, dynamically updating the structural parameters (e.g., load history, environmental data) to enhance bridge safety [117]. This framework takes into account various structural parameters, such as vibration modes, load history, and environmental data, which are dynamically updated through deep learning models to provide real-time analysis and evaluation of the bridge’s condition. However, these models often have high computational costs, especially for deep learning-based frameworks, where the accuracy of the predictions is contingent on extensive training datasets, which can challenge their real-time performance. Moreover, the integration of diverse data sources and systems from multiple disciplines often presents significant challenges in terms of data consistency and interoperability, limiting the effectiveness of DT systems in real-world applications.

Similarly, a DT framework for civil engineering structures, utilizing a dynamic Bayesian network to model the coupling between assets and their digital twins, has been established. This framework incorporates probabilistic graphical models to encode time-varying observations and decision processes, accounting for uncertainty. Real-time structural health diagnosis is achieved through data assimilation using deep learning models, continuously updating the DT state [118]. Furthermore, a DT model for bridge engineering has been proposed, addressing degradation due to material properties, mechanical loads, and environmental factors. By supporting structural analysis and simulating “what-if” scenarios, this model aids in making more reliable maintenance decisions [119].

DT technology has demonstrated substantial potential in bridge structural health monitoring, offering innovative solutions for aging and deteriorating bridge assets. For instance, an intelligent maintenance approach for aging bridge hangers, developed with DT technology, overcomes the traditional limitations in hanger aging and damage detection. By integrating real-time monitoring data, historical maintenance records, environmental stress factors, and dynamic usage patterns, this DT model enables accurate assessments and predictions of hanger conditions [115]. Similarly, an artificial intelligence of things (AIoT)-driven DT communication framework optimizes data synchronization, fault tolerance, and predictive load response in bridge operation and maintenance. This framework leverages edge computing, information hierarchies, and bidirectional communication mechanisms to reduce communication complexity and time delays, demonstrating high efficiency in communication-constrained environments [120]. However, while this framework improves the efficiency, managing large-scale real-time data remains a computational challenge for achieving full operational potential. Scalability in large-scale infrastructure systems is also an ongoing issue, as real-time data processing at such scales can exceed the current computational capabilities.

In addition, a DT framework combining ground-penetrating radar (GPR) scanning for bridge deterioration detection with augmented reality (AR) visualization has been developed. This integration illustrates the trend of combining DT with advanced sensing and visualization tools for effective diagnostics and maintenance. The AR system, deployable on tablets or mixed-reality headsets, integrates visual and location-tracking capabilities with GPR evaluation, validated through field experiments to provide timely preventive maintenance insights [121]. Moreover, an immersive bridge DT platform has been introduced to enhance the efficiency and accuracy of bridge health monitoring and maintenance. The research framework is shown in Figure 7. Integrating BIM, IoT sensor data, virtual reality (VR), AR technologies, and real-time structural feedback loops, the platform addresses issues such as insufficient data visualization and interactivity in traditional systems. By creating a high-fidelity digital replica of a bridge and synchronizing its real-time physical state, engineers can monitor the bridge’s health, diagnose faults, and perform predictive maintenance in a virtual environment [122].

DT technology shows considerable potential in maintenance management by replicating physical asset responses in their corresponding digital counterparts, providing critical insights for informed decision-making [119]. The ongoing evolution of DT, integrated with diverse technologies, is shaping the future of structural health monitoring. The studies highlighted here demonstrate DT’s broad applications in SHM, particularly for bridges, improving monitoring accuracy, enabling precise fault diagnosis, and advancing predictive maintenance. As these technologies evolve, DT is expected to play an increasingly pivotal role in future structural maintenance management, becoming an indispensable tool for intelligent maintenance and resource optimization.

Table 4. Applications of digital twin in maintenance management and decision support.

References	Methodologies
Kang et al. [10]	Intelligent algorithms for creating DT models from large datasets
Shim et al. [21]	DT system integrating 3D models, digital inspection, and image processing for prestressed concrete bridges
Jeon et al. [111]	DT model for bridges incorporating predictive maintenance
Fawad et al. [113]	Integration of IoT, BIM, FEA, and field tests
Hu et al. [114]	BIM-based DT with sensors and wireless communication for data transmission and visualization
Heng et al. [115]	DT-based framework for corrosion–fatigue maintenance in bridge hangers
Yoon et al. [117]	SDeep learning-based framework for integrating 3D modeling and DT for bridge safety
Xu et al. [118]	Dynamic Bayesian network-based DT for asset-coupling and real-time diagnosis
Franciosi et al. [119]	DT model simulating degradation in bridges to support decision-making
Heng et al. [115]	DT model for intelligent maintenance of aging bridge hangers using monitoring and historical data
Gao et al. [120]	AIoT-driven DT framework for optimizing data synchronization in bridge maintenance
Hu et al. [121]	DT framework combining GPR scanning and AR for bridge maintenance
Fawad et al. [122]	Immersive DT platform integrating BIM, IoT, VR, and AR for bridge monitoring

summarizes the studies discussed above.

5. Conclusions

SHM is critical for ensuring infrastructure safety, extending its service life, and reducing maintenance costs. As digital technologies continue to evolve, the integration of DT technology into SHM has garnered increasing attention in recent years. DT enables the creation of virtual models of physical structures, providing a real-time reflection of their dynamic state and offering precise data support for health monitoring throughout their lifecycle [123]. DT addresses traditional limitations such as data acquisition delays and model inaccuracy, enabling comprehensive dynamic assessments.

This paper presents a systematic review of DT application in SHM across three key areas: structural damage detection and identification, dynamic response monitoring and analysis, and DT-based maintenance management and decision support. Research indicates that DT not only enhances monitoring data accuracy and timeliness but also provides robust support for predictive maintenance and fault detection. An in-depth evaluation demonstrated that DT systems significantly reduce false positives in damage detection, improve the real-time adaptability of the monitoring systems, and optimize maintenance scheduling. Notably, DT applications in critical infrastructure, such as bridges and ports, showcase substantial potential, driving continuous innovation and refinement in SHM technologies.

As technology continues to advance, DT applications in SHM are expected to become more intelligent and automated. The integration of drones, robots, and sensors will enable automated data collection, allowing DT to generate detailed and accurate 3D models. This automation will enhance SHM scalability, enabling systems to handle complex structures, reduce safety risks, improve monitoring precision, and expand coverage. These advancements will further support efficient large-scale infrastructure monitoring.

Furthermore, interdisciplinary applications of DT in data processing, machine learning (ML), artificial intelligence (AI), and augmented reality (AR), will be a critical research focus. Fostering collaboration across disciplines such as engineering, computer science, and environmental science can significantly improve the accuracy, reliability, and predictive power of DT models. With growing emphasis on sustainability, the green application of DT is gaining importance. Researchers should focus on minimizing environmental impacts in DT development, optimizing maintenance cycles, improving resource efficiency, and reducing carbon footprints in data processing and model training.

Looking ahead, integrating AR with ML will enhance SHM systems’ intelligence, supporting real-time decision-making for engineers. Advanced ML algorithms can predict structural health, optimize inspection schedules, and reduce maintenance costs through improved decision-making. By analyzing large datasets, ML models can uncover hidden patterns, improving system reliability and resource management in structural maintenance. In conclusion, DT holds transformative potential for SHM, improving monitoring efficiency, reducing costs, and extending infrastructure service life. As the technology evolves, DT will play a central role in the future of SHM. Future research should focus on breakthroughs in model optimization, data fusion, and intelligent decision support, driving further advancements in SHM technologies. Moreover, the integration of cross-disciplinary innovations will be key to addressing the current limitations and enabling future advancements, ensuring that DT continues to shape the future of infrastructure management in a more sustainable, efficient, and precise manner.