Urban Viaduct Structural Health Monitoring: A Review of Wireless Sensor Approaches

Abstract

Urban viaducts have been extensively built in large cities due to traffic pressure on main roads. Although urban viaducts provide relief and reduce traffic, they also pose potential threats to citizens’ safety due to the possibility of collapse. However, most collapses do not happen instantly. Structural health monitoring (SHM) systems can effectively detect structural damage to prevent collapses; however, they are not widely deployed due to the high cost of sensor installation. With wireless sensor technology thriving due to its accuracy and cost-efficiency, the wireless monitoring of massive urban viaducts has become possible. In this paper, we review the full-chain methods of employing wireless sensor technology for urban viaduct structural health monitoring. By proposing a general wireless structural health monitoring approach as a framework, we first review the sensor types and wireless transmission technologies, then the data processing and modal identification methods, and finally the damage detection methods. This paper aims to provide a full-chain review of the mainstream methods in wireless sensor approaches for the SHM of urban viaducts.

1. Introduction

Population expansion in large cities has significantly increased traffic pressure, necessitating the construction of urban viaducts to relieve congestion, as shown in Figure 1. Urban viaducts offer important transportation infrastructure, but they also pose threats to public safety, mainly due to their possible collapse caused by inefficient structural health monitoring (SHM) [1]. Ensuring the serviceability and safety of urban viaducts is critical for people’s livelihoods, as well as the stability and progress of the smart city vision [2].

SHM involves the continuous or regular assessment of a structure’s condition throughout its lifetime [3] to assess its health and performance, as well as identify possible damage or deterioration [1,4]. SHM requires the installation of a network of sensors strategically positioned across the entire viaduct to collect real-time data on structural responses to environmental pressures, traffic, or other external effects and establish a comprehensive view of its health and performance.

It also requires continuous data collection and analysis, which frequently involve the use of modern algorithms and machine learning techniques [5,6] to identify small changes in structural responses. However, the high cost of sensor installation, along with data transmission difficulties, has become a major obstacle to effective and efficient SHM for urban viaducts in recent decades. Advancements in downsizing and integration have resulted in the development of compact wireless sensors with enhanced capabilities. This has made wireless sensors more affordable for large-scale urban viaduct deployments.

On the other hand, the energy efficiency of existing wireless sensors, along with optimized power management strategies, has significantly extended the lifespan of wireless sensors. Efficient wireless communication protocols such as LoRa [7,8] provide reliable transmission of data while also minimizing energy consumption and mitigating the requirement for extensive signal repeaters [2]. The integration of data analytics with cloud computing has resulted in enhanced efficiency in data processing and analysis, leading to a decrease in computational requirements for individual sensors and a reduction in expenses. The combination of these new developments has made wireless sensors easier to install and has resulted in fewer disturbances during operations.

Continuous wireless sensor monitoring can detect various changes or anomalies in urban viaducts through different structural parameters [9]; for example, vibration or excessive movement of the viaduct, structural integrity affected by temperature and humidity, and even cracks that require repair or maintenance. Wireless sensor-enhanced SHM [10] is expected to significantly improve the structural health condition of urban viaducts. This step forward may reduce or completely prevent the possibility of urban viaduct collapses, improving safety.

Wireless sensor technology is thriving, offering a promising prospect for improving the monitoring and maintenance of urban viaducts. However, the wide variety of sensor types, deployment methodologies, analysis approaches, and practical obstacles pose difficulties in establishing practical SHM in actual deployments [11].

This study provides a full-chain review of the state-of-the-art methods in wireless sensor SHM approaches for urban viaducts. Our contributions are summarized as follows:

• We present a general wireless sensor approach (see Figure 2) for SHM in urban viaducts as a framework and further provide a full-chain review of the fundamental sensor types and wireless protocols for sensor data transmission, as well as the data processing, modal identification, and damage detection methods.
• We provide a comparative review, illustrating the differences in the functions and selection of sensors and wireless transmission technologies and highlighting the mainstream SHM structural health assessment methods in the frequency and time domains, with suggestions for deployments.
• We review the state-of-the-art SHM structural damage detection methods, where we compare the representative methods in terms of modal parameters and signal processing. We also review the possibility of integration with machine learning. We further examine the pros and cons of vibration-based SHM methods, with a focus on forced vibration testing and ambient vibration testing. Finally, we outline the future direction of wireless sensor SHM approaches.

This study will provide insights for professionals in both architecture and engineering who are involved in the development of efficient wireless sensor-based structural health monitoring. The rest of this paper is organized as follows. Section 2 reviews SHM data collection, including wireless sensors and data transmission. Section 3 reviews the SHM data processing and modal identification methods. Section 4 focuses on reviewing SHM structural damage detection. Section 5 provides a comparative review of vibration-based SHM methods. Section 6 provides our concluding remarks and outlines our future work.

2. SHM Data Collection

In this section, we describe the data collection for SHM in urban viaducts and review the functions and selection of sensor types and wireless data transmission technologies.

2.1. Sensor Types

Sensors are summarized in this subsection, in which we focus on those for maintaining the SHM of viaducts. We also discuss the factors involved in sensor selection, including compatibility with wireless networks, power consumption, and suitability for detection algorithms.

2.1.1. Accelerometers

An accelerometer is a sensor that monitors the rate of change in the vibration of an object or structure [12]. A vibration of a viaduct may occur due to dynamic parameters, including traffic movement, wind forces, or seismic activity [13,14,15]. Structural damage can be detected by monitoring the vibrations, since changes in the bridge’s stiffness or mass can be identified by detection methods [16,17]. The terms viaduct and bridge are used interchangeably in this review.

The detection of vibrations with low amplitude and low frequency is necessary for their usage in applications related to SHM. Piezoelectric accelerometers and Micro-Electro-Mechanical System (MEMS) accelerometers [18] are two types of accelerometers employed in concrete structures for identifying vibration and excitation through dynamic parameters [19].

Current state-of-the-art MEMS accelerometer sensors have demonstrated high measurement accuracy compared to classic wired or wireless piezoelectric accelerometers [18]. M.J. Whelan et al. [13] proposed a MEMS accelerometer (see Figure 3a) and a Bridge Diagnostic Index strain transducer. Structural behavior under ambient and traffic loads can be assessed using accelerometers to investigate the natural frequencies, damping ratios, and mode shapes of the viaducts. Structural excitation is primarily caused by low ambient environmental stresses and traffic. The transmission of data is completed with minimal loss. The validation of the proprietary wireless sensor approach in an operational service context is confirmed through the successful acquisition of high-rate, lossless data on highway bridges.

Huynh et al. [20] deployed a MEMS accelerometer on the Hwamyung Bridge, a cable-stayed bridge located in Busan, South Korea. The accelerometers were strategically positioned on several components of the bridge, including the deck, pylon, and stay cables. The sensors obtained data pertaining to the acceleration, velocity, and displacement of the bridge. The results of the investigation indicated that the method effectively determined the structural parameters of the bridge, including its natural frequencies, mode shapes, and damping ratios. The method demonstrated its ability to identify alterations in structural characteristics resulting from typhoon events.

Figure 3. Examples of sensors used in the literature for SHM: (a) MEMS accelerometer [13], (b) vibrating wire sensor [21], and (c) MEMS inclinometer [22].

Figure 3. Examples of sensors used in the literature for SHM: (a) MEMS accelerometer [13], (b) vibrating wire sensor [21], and (c) MEMS inclinometer [22].

2.1.2. Strain Sensors

Strain sensors can detect structural damage in viaducts and are generally positioned at critical areas on bridge structures to measure stress and strain levels.

When a load is applied to a viaduct, strain sensors detect the ensuing deformation and transmit the information to a monitoring system. This information can be utilized to arrange maintenance activities in advance of the damage becoming serious [23]. Embedded stress sensors and piezoelectric sensors are two types of strain sensors commonly used for monitoring concrete-based structure health.

Piezoelectric sensors, also known as cement-based strain sensors, have proven useful in the health monitoring of concrete structures and can be embedded in structures [24]. Vibrating wire (VW) sensors (see Figure 3b) were employed by Yanbin Shen et al. [21] for deployment at the Olympic Stadium of Beijing, China. They can collect long-term strain measurements due to their stability, durability, and resistance to electromagnetic components, making them suitable for steel and concrete components.

After attaching a VW sensor to the surface of a measured component, the wires in a VW sensor vibrate and experience fluctuations in the frequency due to tension and compression. Note that a plucking and pickup coil is responsible for exciting the wires and capturing the resonance frequency. A temperature sensor is also integrated into VW sensors, since temperature is one of the critical factors that affect strain measurements [25]. To this end, Sahar Hassani et al. [26] proposed a temperature compensation method that mitigates the temperature effect on strain measurements.

2.1.3. Inclinometers

An inclinometer is a sensor used to measure the angle of inclination of a structure. This sensor can be applied to measure the deflection of a viaduct by monitoring the change in the angle of inclination of the viaduct deck.

Yan Yu et al. [22] employed a MEMS inclinometer, as shown in Figure 3c, to measure the deflection of a viaduct by measuring the change in the angle of inclination, which was achieved by installing an inclinometer on the viaduct deck and positioning it at the location expected to reveal the highest degree of deflection. Note that the inclinometer was calibrated to ensure that the reading displayed a value of zero.

When a bridge undergoes deflection, a pendulum or tilt sensor undergoes rotation, and the readout device displays the magnitude of deflection. The measurement of viaduct deflection can also be established by utilizing a combination of inclinometers. A pair of inclinometers can be mounted to the bridge deck, positioned at opposite ends of the viaduct. The difference in the readings obtained from the two inclinometers is equivalent to the amount of deflection of the viaduct [22,27,28,29].

A tri-axis accelerometer can also be employed to enable the determination of the inclination of an object or civil structure based on the measurement of its acceleration along three spatial axes. The gravitational force, which indicates a continuous direction toward the Earth’s center, influences the acceleration of the object or structure. Therefore, by measuring the structure’s acceleration along the three spatial axes, the object’s or structure’s inclination angle in relation to the horizontal plane can be determined.

To determine the inclination of an object or civil structure using a tri-axis accelerometer, the accelerometer must be attached to the object or structure in a position that ensures its alignment with the horizontal plane. The acceleration measurement in the three spatial axes of an object or structure can be conducted once the accelerometer is properly positioned.

Recently, a diverse array of sensors has become available, utilizing various technologies, each with a wide range of capabilities to support SHM. Among the various options, integrating an accelerometer with vibration-based techniques has emerged as a promising approach for SHM [30].

Overall, sensors offer the capability to detect minor changes in the dynamic response of viaducts, potentially indicating signs of damage or deterioration. Sensor data can provide deep insights into the status of a viaduct, enabling early warnings, maintenance decisions, and safety procedures. However, note that environmental factors may affect sensor readings and further affect modal identification: temperature and humidity can cause hardware variations such as frequency drifting [31], which can lead to data fluctuations, while heavy operational loads can also cause hardware thermal variations.

Moreover, the novel sensor fusion strategy, which combines multiple sensor data, can further enhance the robustness and accuracy of SHM. Ref. [32] integrated sensor readings from distributed devices and converted these readings into actionable information for SHM. Ref. [33] proposed multi-sensor data fusion to enhance the consistency and robustness of monitoring systems. Ref. [34] explored the adoption of an advanced encoder transformer for handling data fusion in SHM. Multi-sensor data fusion is a promising approach for achieving robust and highly accurate SHM.

2.2. Wireless Sensor Data Transmission

Wireless networks employed for SHM sensor data transmission can be categorized into four main types: Wireless Personal Area Networks (WPANs), Wireless Local Area Networks (WLANs), Wireless Neighborhood Area Networks (WNANs), and Wireless Wide-Area Networks (WWANs) [35].

2.2.1. WPANs

WPANs are designed for short-range communication, where transmissions are often restricted to a few meters. WPANs are typically used for short-range data transmission within a group of devices and require minimal infrastructure.

Wireless communication protocols in WPANs, including Radio-Frequency Identification (RFID), Bluetooth, and Zigbee [36], are widely employed. RFID is a cutting-edge technology that utilizes radio waves to precisely identify and monitor various sensor data; it can be employed for diverse purposes, such as monitoring inventories, managing access control systems, and identifying locations for viaduct SHM.

On the other hand, Bluetooth is a widely used technology that has multiple purposes for SHM in viaducts, such as sensor data streaming. ZigBee is a wireless technology specifically developed for wireless sensor and control networks that demand low power consumption, low cost, and low data rates. Note that ZigBee is extensively deployed in multi-hop viaduct SHM.

2.2.2. WLANs

WLANs are a group of wireless protocols designed to efficiently transmit data between devices within a range of tens to hundreds of meters. Mesh WLANs, a more advanced generation of ordinary WLANs, employ several access points to expand network coverage and improve performance in larger regions, ensuring uninterrupted connectivity across expansive locations.

The coverage of WLANs is greater than that of WPANs but significantly smaller than that of wide-area wireless communications; thus, the employment of WLANs in viaduct SHM is primarily suitable for medium-scale viaducts.

2.2.3. WWANs

WWANs have a greater coverage area than WLANs. For example, cellular networks such as 4G and 5G are built to facilitate the rapid transmission of data over long distances, ranging from a few to tens of kilometers. However, data transmission through cellular networks is quite expensive in terms of hardware and energy costs.

A special category of WWANs is the Low-Power Wide-Area Network (LPWAN) [37], which is tailored for long-distance communication and requires low power consumption at the expense of limited bandwidth. LPWANs have diverse implementations, and in

Table 1. Comparison of LoRaWAN, Sigfox, and NB-IoT [37].

Feature	LoRaWAN	Sigfox	NB-IoT
Modulation	Chirp	GFSK/UNB	OFDM
Data Rate	290 bps–50 kbps	100 bps	100 kbps
Battery Lifetime	Years	Years	Years
Power Efficiency	Low	Low	High
Range	2–5 km urban	3–10 km urban	3 km urban

, we summarize some of them, including LoRaWAN, Sigfox, and NB-IoT [37,38,39]. LPWANs are an ideal solution for SHM of large-scale viaducts, and their low-power, wide-area transmission is essential for achieving real-time, long-term, and continuous SHM of urban viaducts.

2.2.4. LPWANs

We chose LoRaWAN from Table 1 to illustrate its advantages for SHM of large-scale urban viaducts due to its long-distance communication ability and low power consumption [40]. LoRaWAN allows long-range communication, often ranging from several kilometers to tens of kilometers.

An example deployment of a LoRaWAN for viaduct SHM is illustrated in Figure 4a. The LoRa nodes are made up of sensors, actuators, batteries, and other essential components. They perform diverse tasks according to different SHM requirements, providing the capability to generate, send, and receive SHM sensor data to and from other nodes and gateways.

In LoRaWANs, gateways are essential components that function as a connection between end nodes and network servers. They transmit packets in both the uplink and downlink directions, guaranteeing sensor data and controlling data transmission. The gateways also perform tasks, such as inspecting bit error rates to guarantee the integrity of the data, managing admissions to control network access, and implementing security measures that protect the data.

LoRaWANs commonly utilize a star topology (see Figure 4b) in their network architecture to achieve low power consumption. In a star topology, each node establishes its connection directly with the gateway, avoiding the need for additional devices to relay data; all its communications with other nodes have to pass through the gateway. Therefore, data transmission in such a topology usually requires only a single-hop communication from the node to the gateway. This centralized structure simplifies the administrative process in SHM.

Moreover, the star topology in LoRaWANs offers the benefit of enabling easy fault detection. Each node in the network is directly connected to the gateway, ensuring that the failure of one node does not cause interruption to the entire network. This further enhances the efficiency of troubleshooting and helps to ensure the reliability and accuracy of data collection in SHM.

In actual deployments, LoRa nodes are distributed across the monitoring scene; thus, their time must be synchronized for accurate data ordering in modal identification in SHM. A number of highly accurate time synchronization methods [8,40] have been proposed for LoRa. The more accurate the time synchronization among the LoRa nodes, the higher the accuracy and validity of the sensor data ordering, and the higher the accuracy of modal identification in the SHM system [41,42].

Of particular note is that, besides the above wireless transmission protocols, there are also several energy-harvesting communications to further enhance the deployment lifetime in the SHM of viaducts. Ref. [43] revealed the feasibility of integrating energy-harvesting WSNs for structural health monitoring. Ref. [44] summarized the energy sources and harvesting techniques for energy harvesting in WSNs. Ref. [45] further presented how energy harvesting can be achieved for wireless sensor nodes in viaduct SHM.

2.2.5. Topology and Scalability

Scalability is a vital aspect in designing SHM for urban viaducts. The star topology is particularly effective in such cases: when there is a need to add new SHM nodes to the network, they can be easily integrated by establishing a connection with the gateway. Achieving scalability in this way is beneficial in situations where the monitored viaducts may experience expansions in the future.

Although the star topology has many benefits, it is important to emphasize its drawbacks, particularly its reliance on the gateway. A major constraint is the limitation on the number of devices that can be connected to the network, which is defined by the gateway’s capacity. The gateway acts as a central hub for all communications in the star topology, and its capacity determines the maximum size the network can achieve. When connected devices reach or exceed the hosting capacity of a gateway, congestion and possible interruptions may be encountered in data transmission.

When conducting SHM of urban viaducts, the failure or damage of a single node has only a minor effect, affecting only that particular node; however, in the case of a breakdown in the gateway, the entire network suffers a total disruption. This single point of vulnerability presents a fatal risk to the reliability of SHM, especially in scenarios where continuous monitoring is required.

In contrast, the mesh topology, as shown in Figure 4c, has been proposed to counter the single-point failure. Each node is interconnected with multiple neighboring nodes, creating a mesh of communication paths. Unlike the star topology, where nodes connect to a gateway or central hub, the mesh topology allows nodes to transmit directly with each other. Each node can directly communicate with other nodes, thus avoiding the need for a central gateway. The mesh topology is considered a more reliable network topology, but it is complex, and its establishment requires a large amount of power and resources, which, unfortunately, contradicts the primary need for energy efficiency in the SHM of urban viaducts.

Moreover, in large SHM coverage involving many long-distance transmissions, deployment faces interoperability issues, while scalability faces major challenges in terms of data loss, noise, and errors. To handle interoperability, cross-technology is often employed, where different wireless protocols can translate their packet contents for communication [46,47]. Regarding scalability, we use LoRa wireless sensor transmission as an example to illustrate how the data loss and error correction are handled in the literature. Ref. [48] studied the path loss model, which guides path loss estimation to reduce data loss in long-distance transmission. Ref. [49] provided an in-depth study on the performance of channel access to alleviate data loss. Ref. [50] investigated the error correction technique for maximizing data availability. Ref. [51] studied the error correction under interferences. Experimental evaluations in [52] further revealed the impact of transmission parameter configuration on packet reception performance.

3. SHM Data Processing and Modal Identification

Modal identification (MI) is a widely employed technique used in mechanical engineering to understand a system’s dynamic behavior independently of the applied loads and the system’s response. Recently, MI has been introduced into viaduct SHM, where its methods of experimental estimation and condition identification through modal parameters are employed to investigate structural health and assess performance. We categorize existing methods that analyze structural data into frequency-domain and time-domain methods and review them in this section.

3.1. Frequency-Domain Method

The frequency-domain method is employed in SHM to examine a structure’s dynamic behavior. This method is especially useful in structural damage detection. The frequency-domain method involves converting the structural responses from the time domain to the frequency domain. Such a transformation enables the examination of the dynamic properties of the structure, including key properties such as natural frequencies, damping ratios, and mode shapes.

Various Experimental Modal Analysis (EMA), or input–output modal analysis, and Operational Modal Analysis (OMA), or output-only modal analysis, techniques are utilized, which estimate the Frequency Response Functions (FRFs) [53] or Impulse Response Functions (IRFs) obtained using Fast Fourier Transform (FFT) techniques.

An FRF quantifies the degree to which the output response of a structure corresponds to the applied force. Furthermore, the FFT algorithm is utilized to convert the output response or observed time-domain data—e.g., displacement, velocity, and acceleration—of a structure from the time domain to the frequency domain [54].

Singh et al. [55] utilized the FRF method to capture the dynamic response of a two-story building model to assess its resilience to external disturbances such as earthquakes. An accelerometer detects vibrations and updates the Finite Element Model (FEM) to forecast behavior and detect damage during seismic occurrences. FRFs are employed to forecast the structural dynamic response under different damage scenarios because of the intricate connection between the input and output signals. The study concluded that natural frequencies are affected by the location and level of damage, resulting in greater frequency reductions in higher modes.

On the other hand, the FFT and other methods were utilized by Amezquita-Sanchez et al. [56] in a civil structure to determine the dynamic response during artificial seismic activity using a shaker. Their results showed that the FFT is an effective tool for understanding the impact of seismic activity on civil infrastructure and assisting in the construction of more resilient structures.

3.1.1. Peak Picking Method (PP)

PP is a straightforward and efficient method employed to determine MI in the frequency-time domain. This method has been adopted for various reasons over the years. The principle of proximity placement states that a structure has notable responses at its natural frequencies when subjected to external stimuli. The peaks in the Power Spectral Densities (PSDs), calculated from the time histories gathered at the measurement locations, can help identify these frequencies.

This method involves determining the natural frequencies by identifying the peaks in the averaged normalized PSDs. The natural frequencies are directly obtained from the PSD plot at the peak. Kotsakos et al. [57] employed PP by deploying an accelerometer and using the FRF to achieve SHM. Then, frequency-domain decomposition was introduced, which is particularly useful when only output data are measured, making it valuable for structural health monitoring and modal analysis.

3.1.2. Frequency-Domain Decomposition (FDD)

FDD is a method typically employed in output-only modal analysis and involves estimating the modal characteristics of a structure only from its response data without requiring knowledge of the input forces. To determine the mode shapes and frequencies, the process involves breaking down the PSD matrix of the structure’s response into singular values and vectors using Singular-Value Decomposition (SVD).

FDD can determine damping by converting the auto-spectral density functions of the identified Single-Degree-of-Freedom (SDOF) systems back to the time domain using the inverse FFT, analyzing free decay, and computing the damping ratio and natural frequencies. FDD is known for its dependability and efficiency in estimating modes, particularly in scenarios involving intricate structures and scarce data on external forces. FDD offers a methodical and verified way to estimate damping by modifying the modal assurance criterion limit and assessing the quality using basic graphs [58,59,60].

3.2. Time-Domain Method

The time-domain method utilizes an SDOF for calculations but processes time-series data. The SVD of the correlation matrix output is employed to isolate the undamped mode shapes related to the sensor locations. PP has also been employed in the time domain to determine the natural frequencies and damping ratios from the SVD signal, following the acquisition of the mode shapes. Multiple unique algorithms are utilized in this method.

3.2.1. Natural Excitation Technique (NExT)

As a theoretical basis for NExT, James et al. [61] showed that random inputs can cause a multiple-input multiple-output, multiple-mode system to produce autocorrelation and cross-correlation functions that look like damped sinusoids [61]. These damped sinusoids must share identical frequencies and decay with the structural modes. The correlation functions have a combination of sinusoids as their modal forms within a smooth modal basis algorithm.

A universal solution for a structure made up of discrete spatial elements should be proposed for using this method. Then, the cross-correlation function between results should be computed, and the situations where the inputs are random should be further handled. The theoretical justification for NExT can be established for a broad range of random inputs, complex modes, and white-noise inputs with known harmonic inputs. This advancement will only apply to exact inputs, actual modes, and no harmonics.

The estimation of modal parameters solely from multiple-channel responses under ambient excitation presents challenges, particularly when handling asynchronous responses due to sensor errors. Yang et al. [62] proposed a novel approach to address this issue, introducing a method based on NExT, where the power spectra of monitoring responses are initially calculated and transformed into correlation functions through the inverse Fourier transform.

By comparing the phase characteristics of power spectra and the linear dependency of modal components in correlation functions for synchronous and asynchronous responses, Yang enhanced NExT. Through a meticulous process involving the minimization of phase slope and maximization of linear dependency of modal components, the enhanced NExT synchronizes multiple-channel responses, aligning them to yield high-precision mode shapes. The efficacy of Yang’s approach was validated through numerical examples and real-world applications, demonstrating its potential to advance operational modal identification in viaduct health monitoring.

3.2.2. Autoregression Moving Average (ARMA)

According to [63], the ARMA model is a time-domain method used in Operational Modal Analysis (OMA) to determine modal parameters from structural responses. This method predicts the present values of a time series by utilizing past values and a prediction error, which is beneficial for analyzing linear systems. The ARMA model expands on the idea of a linear time-invariant system driven by white noise, under the assumption that the observed output is stationary.

Vector ARMA models are used when there are many input excitations. The prediction error is commonly used with ARMA models to determine modal parameters by reducing the discrepancy between the estimated response from the model and the actual measured response. Despite its effectiveness, the ARMA technique has limitations: it can be computationally intensive and may not always reach convergence in every scenario.

ARMA methods have been employed in civil engineering applications in the past, but their popularity has decreased because they require a large amount of computation time. Several ARMA methods, including the instrumental variable, linear multi-stage, and two-stage least squares techniques, have been created, all of which require significant computational resources.

3.2.3. Stochastic Subspace Identification (SSI)

According to [64], SSI methods can be categorized into two primary groups: covariance-driven (Cov-SSI) and data-driven (DD-SSI). Cov-SSI utilizes the covariance matrices of the measured data to detect system modes, whereas DD-SSI directly analyzes the observed data to create a data matrix. SVD is a crucial component of both Cov-SSI and DD-SSI; it decomposes the data matrix into its singular vectors and singular values, which correspond to the fundamental spatial modes and modal frequencies of the structure.

SVD results allow for the extraction of modal data for SHM, including natural frequencies, mode shapes, and damping ratios. Chang et al. [65] employed and validated the application of SSI. SSI must allocate computational resources to process a substantial amount of data. The proposed order selection criteria were used to establish the number of structural modes, which, in turn, led to the determination of the structure’s MI.

SSI has also been employed to analyze the practicality of a real-time application, where it was applied to an actual viaduct [66]. The results demonstrated that the enhanced SSI effectively reduced computational time while maintaining qualitative modal parameters. We summarize the pros and cons of all of these methods in

Table 2. Summary of modal identification methods.

Method	Type	Advantages	Disadvantages
PP [57]	Frequency Domain	The most basic, direct, and computationally efficient method	Not applicable to systems with closely spaced modes, which is the case in most real structures
FDD [58]	Frequency Domain	Precise identification of natural frequencies and mode shapes that are tightly spaced is possible	Calculating damping ratios is not possible
NExT [61]	Time Domain	Robust basic structural dynamic properties	Relies on structured, consistent processes.
ARMA [63]	Time Domain	Easy setup	Extensive computations needed
SSI [65]	Time Domain	High level of parameter estimation accuracy	Mathematically complex method

.

In summary, various factors should be considered when selecting a processing method in a practical SHM application. First, the selection should be based on the computational resources of the underlying processing platform. The computational complexity of PP is the lowest among all the methods, that of SSI is the highest, and that of FDD, NExT, and ARMA is moderate. Second, the selection should consider the application scenario. PP is suitable for rapid screening and preliminary modal identification, FDD is suitable for multi-modal separation and offline analysis, NExT is suitable for non-stationary data preprocessing, ARMA is suitable for online processing of short data segments, and SSI is suitable for detailed modeling of complex structures. Note that, in practice, the actual selection is more complicated and should jointly consider many factors such as sensor data, deployment requirements, platform, and environment. The processing methods are compared in Table 2.

4. SHM Structural Damage Detection

As illustrated in Figure 5, structural damage detection methods in SHM are commonly classified into two categories: global methods and local methods. Note that local SHM methods aim to identify and assess structural damage on a smaller scale, without relying on or analyzing the structural vibration response.

The classification of most Non-Destructive Testing and Evaluation (NDTE) methods as local is primarily due to their limited detection range. Ultrasonic testing, acoustic emissions, infrared thermography, visual testing, and various other techniques are employed to examine, assess, and appraise structural components and assemblies within local areas of infrastructure [67,68,69,70]. In addition, the expertise of a trained technician is required to conduct the examination, establish threshold values for signal processing, and interpret the results.

Monitoring large structures using traditional NDTE methods is often costly and time-consuming. A review of NDTE methods and other local methods for damage detection, localization, and quantification is not included in this paper. This section exclusively focuses on vibration-based global SHM techniques. Nevertheless, it is crucial to acknowledge that local techniques are hardly suitable for the long-term SHM of large structures like viaducts and high-rise buildings. Consequently, global SHM methods are discussed due to the size and complexity of these structures.

Vibration-based techniques are popular and reliable in SHM and include four main steps. Due to their non-destructive nature, which allows for continuous long-term monitoring without process interruptions, these methods have gained popularity and demonstrated reliability in SHM in most industries. Vibration-based techniques can acquire structural dynamic parameters such as natural frequencies, mode shapes, and damping ratios from the dynamic response of a structure. These structural dynamic parameters, also known as modal parameters, can change due to external excitation, which may indicate structural damage or degradation. The dynamic response of a structure can be monitored using sensors such as accelerometers, strain sensors, or fiber optic sensors, but this review focuses on accelerometers due to their low power and reliability, as mentioned in Section 2.1.

On the other hand, tri-axis acceleration monitoring is a dedicated methodology that involves the simultaneous analysis of vibrations along the X-, Y-, and Z-axes. Vibration-based techniques offer a comprehensive understanding of how a structure reacts to external forces or dynamic forces such as wind, traffic loads, and earthquakes. This dynamic excitation will induce a dynamic response in the structure, which is essentially for identifying modal parameters [71,72,73,74].

The extraction of features from collected vibration signals is a challenging task, as it is influenced by the presence of noise [75]. The construction of a complex data processing and analysis system is crucial for obtaining precise data related to the health of the structure. By comparing the healthy dynamic response of a structure to the damaged dynamic response, vibration-based methods can detect damage to the structure. Changes in the dynamic response of a structural parameter, captured with respect to an established reference state, can easily be used as indicators of structural damage, damage location, and damage severity [76].

Comprehensive research has been conducted on structural damage detection in viaduct SHM, which relies on vibration measurements from viaducts. These studies aim to identify variations in the dynamic response parameters, including the natural frequency, modal strain energy, mode form curvature, and dynamic flexibility [77]. For instance, a decrease in natural frequencies indicates structural degradation or damage caused by an intense event, which leads to a decrease in stiffness [78].

According to [77], natural frequencies are considered highly significant indicators in vibration-based SHM methods for detecting damage in a structure. A large number of research studies have reported on the use of natural frequency characteristics in the deployment of vibration-based damage detection systems for viaduct structures. Hou et al. [78] defined structural damage as alterations in the parameters of structural characteristics, such as the dynamic response parameters, that harm the structural integrity of the system. This includes changes in the mass, damping, and stiffness of the structure. Therefore, an in-depth review is necessary.

4.1. Natural Frequencies

Magalhaes et al. [79] employed a vibration-based SHM method to analyze the changes in dynamic parameters, particularly the natural frequencies, of a concrete arch bridge. Their research focused on monitoring the first 12 natural frequencies of the bridge over a duration exceeding two years. They aimed to reduce the impact of environmental and operational factors on the inherent frequencies of the bridge, thereby facilitating the precise detection of damage. In their research, the ability of a monitoring system to detect anomalous behavior was influenced by the amount of data collected before the occurrence of the unusual event. Note that a larger dataset can result in a more accurate determination of reference values.

Further, Fan et al. [80] studied prestressed concrete viaducts. Extensive experiments were conducted on highway viaducts to identify and measure damage by analyzing alterations in natural frequencies and mode shapes. Empirical observations demonstrated that alterations in dynamic properties can be used to identify damage in its initial stages. Nevertheless, natural frequencies alone are insufficient to accurately measure the extent and specific area of damage to the structure.

4.2. Finite Element Model

The FEM is a computational tool employed in SHM to model and examine the response of complex structures to different excitation sources. This method involves breaking down a complex system into smaller and more manageable components, known as finite elements, and material characteristics, geometric properties, and boundary conditions are assigned to each element, enabling the accurate estimation of the structural response to external excitation sources. FEM is employed to build a model of a structure—e.g., a viaduct or building—that accurately captures its physical characteristics, including its mass, damping, and stiffness. Afterward, the model is used to forecast the behavior of the structure in different scenarios.

Over time, a structure may experience alterations due to aging, climatic conditions, or damage. These alterations can lead to changes in the structure’s physical characteristics from those indicated in the original FEM analysis. To address this, FEM is revised using recorded vibration responses, which is referred to as model updating. By updating the FEM, it can be compared to the original model of the structure that has not been damaged. Any significant differences between the updated model and the reference model may suggest the possibility of damage. This approach serves as a means of monitoring and detecting deterioration [81].

Cheng et al. [82] conducted experiments on large transmission towers in China. These towers exhibit high sensitivity to environmental pressures, which showcases the importance of ensuring their safety. The researchers created a FEM-based SHM method for a transmission tower that is 131 m tall. The process involved detailed modeling, manual tuning, model updating using the response surface method, and validation. Before the installation of sensors, in situ measurements were taken to determine the dynamic characteristics, yielding crucial data for the FEM. By manually adjusting and updating the model, the researchers further enhanced the accuracy of the FEM. A seismic analysis demonstrated that model updating accurately represented the tower’s seismic performance, resulting in alterations in displacement and stress distribution.

Table 3. State-of-the-art vibration-based SHM methods.

References	Modal Parameter	Signal Processing	Description
Alamdari et al. (2016) [83]	Natural Frequency	FRF	Sydney Harbor Bridge
Khatir et al. (2018) [15]	Natural Frequency	FRF	Beam-like structure constructed from steel
Yang D et al. (2010) [84]	Natural Frequency, Displacement, Temperature	ARMA	Runyang Suspension Bridge in China
Alves et al. (2021) [85]	Natural Frequency	SSI	Tianjin Yonghe Bridge
Magalhães et al. (2010) [79]	Natural Frequency, Temperature	FDD, SSI	Concrete bridge
Saidin et al. (2023) [86]	Natural Frequency, Mode Shape	FDD, SSI	Ultra-high-performance concrete bridge
Cabboi et al. (2014) [87]	Natural Frequency, Mode Shape	FDD, SSI-COV	San Michele iron arch bridge
Cheng et al. (2018) [82]	Natural Frequency, Mass, Damping, and Stiffness	FEM	Transmission tower in China
Fan et al. (2011) [80]	Natural Frequency, Mode Shape	FDD	Prestressed concrete bridges
Wang et al. [77]	Natural Frequency, Mode Shape	FFT	Victoria Bridge
Chang et al. [65]	Natural Frequency	SSI	Cable-stayed bridge in Taiwan
Kotsakos et al. [57]	Natural Frequency	PP, FRF, FFT	Multi-layer buildings
Mezquita-Sanchez et al. [56]	Natural Frequency	FFT	Cube-shaped structure

presents a summary of vibration-based SHM methods. Note that the modal parameters of a structure, especially its natural frequencies, possess high sensitivity to even minor damage. Besides the conventional methods discussed above, the advent of machine learning has enabled significant improvements in modal identification and damage detection. Ref. [1] provided a comprehensive review of the effectiveness of machine learning in damage clustering, regression, and classification for diverse structures. State-of-the-art research has further expanded the application of machine learning in damage detection. Ref. [5] employed machine learning to detect structural energy time-history responses for damage detection, while [6] designed a hybrid transfer learning-based neural network to detect damage in bridges. With the development of new artificial intelligence methods, structural damage detection solutions will continue to flourish.

5. Vibration-Based SHM Methods: Pros and Cons

Having reviewed the state-of-the-art vibration-based SHM methods, we provide more insights into two of these methods in this section. It is worth mentioning that vibration testing for dynamic analysis in SHM involves the process of regularly evaluating a structure’s dynamic response to external forces or operational conditions. Excitations can be applied as forced or ambient vibrations to observe the dynamic response of a structure and obtain modal parameters, such as natural frequencies and mode shapes.

Although the vibration-based methods described below utilize different input excitation methods, they are two common approaches used in dynamic analysis to evaluate the health of structures such as viaducts, buildings, dams, and pipelines.

5.1. Forced Vibration Testing (FVT)

FVT employs a controlled external excitation or dynamic load to a structure at particular frequencies or amplitude levels. This excitation can be produced by mechanical shakers, electromagnetic actuators, impulse hammers, or other controlled artificial vibration sources. Conversely, ambient vibration testing monitors a structure’s dynamic response caused by external uncontrolled excitation like wind, traffic, or machinery. Ambient vibration testing utilizes the natural vibrations present in a structure’s environment during regular operation, unlike forced vibration testing, which requires controlled excitation sources. Both methods aim to efficiently detect anomalies or damage to prevent major breakdowns and ensure structural safety and reliability.

Existing research has investigated the effectiveness of FVT in detecting structural damage by analyzing how structures respond dynamically to controlled input excitation. These methods include applying a controlled external artificial excitation to the structure and then analyzing the dynamic response to obtain important information about the structural health and integrity of the structure. During FVT, researchers can adjust the input excitation parameters, including the frequency, amplitude, and duration, to meet the experiment’s specific requirements.

Structural health can be assessed by comparing the modal parameters of undamaged structures with those of damaged structures. Chen et al. [88] applied a static load using a 30-ton hydraulic jack equipped with a donut-shaped load cell. Following each static level of stress, the load apparatus was taken away to allow room for the dynamic test setup. A full-scale beam with fiber-reinforced polymer sheets was tested in the study. It was shown that the natural frequencies of the beam were lower at the start of cracking and then stabilized. The unstrengthened beam’s frequency decreased continuously as it underwent concrete cracking and reinforcement yielding, but the proposed technique may not be suitable for assessing significant cracks or extensive damage that has persisted over an extended period.

Orhan et al. [89] employed the FEM of a beam to simulate a crack on the beam. The study revealed that the natural frequencies of a beam decreased with constant crack location but increased with crack depth. The dominant response occurred at 96 Hz for a single crack on the top and bottom surfaces, while additional frequencies appeared for two cracks on the top and bottom surfaces. The study revealed that degradation or damage can be detected using the FVT method by examining the natural frequencies of the structure.

However, this method cannot pinpoint the exact location of the damage. A full-scale synthetic two-story high-performance building was tested under ambient and forced vibrations in a controlled laboratory environment by Lamarche et al. [90]. The building was subjected to simulated artificial earthquake damage for forced vibration testing, and ambient vibration testing was also performed with human walking as an input excitation. In the study, FVT was regarded as a reliable method for experimentally evaluating the modal parameters of a structure or building.

Table 4. Pros and cons of forced vibration testing.

Advantages

Disadvantages

Customizes EMA by adjusting input excitation parameters to target various aspects of structural health. Precise damage detection, such as cracks and material degradation. Fast and straightforward approach to excite the structure.

Not suitable for assessing significant cracks or extensive damage that has persisted over an extended period. Does not provide precise details about the exact location of the damage within the structure. Specialized equipment is often complex and costly. Improperly conducted forced vibration testing may lead to increased stress or harm to the structure. Testing may interfere with regular operations, particularly for structures that are already in use.

illustrates the pros and cons of FVT.

FVT has been applied to in situ structures in several studies [91,92,93], showing that the dynamic responses of the structures were accurate.

5.2. Ambient Vibration Testing (AVT)

As the equipment required for forced excitation has become cumbersome and costly, employing natural excitation sources such as wind and waves has led to the development of alternative techniques for MI. AVT has emerged as a practical solution, which avoids the need for mechanical excitation devices and is particularly beneficial for large structural bridge applications.

Alamdari et al. [83] proposed a spectral-based clustering technique that aims to find the characteristics in the FRFs of the Sydney Harbor Bridge structure that were damaged. The authors emulated several damage scenarios by selectively removing cables and hangers from the bridge and then monitoring and extracting the dynamic responses of the bridge under uncontrolled traffic loading. Accelerometers placed on the bridge deck and arch employed the wavelet transform to extract spectral features, which were used to determine the natural frequencies of the structure. K-means clustering was then applied to group these spectral features. The results showed that the location of damage can be effectively identified and localized.

Saidin et al. [86] also proposed a method using AVT for the SHM of an ultra-high-performance concrete (UHPC) bridge to acquire the natural frequencies and mode shapes. They compared the modal parameters with the FEM of the bridge and used simple linear regression to evaluate and predict the bridge’s condition. Their results showed that the vibration-based method was effective and reliable for detecting and evaluating the structural condition of the UHPC bridge.

Alves et al. [85] proposed an SHM system for the Tianjin Yonghe Bridge, where the acceleration data were normalized to eliminate the effects of scaling and offset. In addition, they applied a wavelet transform to break down the data into different frequency ranges and identified those that were relevant for damage identification. They also extracted the natural frequencies of the structure, which were utilized to indicate the precise locations of structural damage with various levels of severity.

Yang et al. [84] proposed an approach that employed extended monitoring data and statistical modeling methods to identify abnormal changes based on natural frequencies and displacement, thus mitigating the impact of temperature-related issues on the Runyang Suspension Bridge in China. The approach included a number of sensors, such as accelerometers, displacement sensors, and temperature sensors. The authors employed a statistical modeling approach to establish the relationship between the natural frequencies, temperature, and displacement of the bridge. However, the steps and methods used to extract the modal frequency of the bridge were not provided. The findings showed that the proposed method can detect anomalous changes in the natural frequencies and displacement of the bridge.

Magalhaes et al. discussed the use of vibration-based SHM approaches for bridges, specifically focusing on modal parameter tracking. Output-only modal analysis was highlighted as a way to identify a structure’s modal parameters under operational conditions, with natural frequencies being commonly used as damage indicators. AVT was employed to identify natural frequencies. Then, regression analysis and principal component analysis were employed to reduce the influence of environmental and operational factors. The proposed method showed that the capability of SHM to detect abnormal behavior depends on the amount of data collected before the occurrence of an event. The time needed to detect damage is inversely proportional to the extent of the damage.

Cabboi et al. [87] conducted research to compare damage detection by obtaining the natural frequencies and mode shapes of the San Michele iron arch bridge. Although no damage was detected in the research, it was determined that using AVT to acquire the modal parameters was successful.

In summary, the advantages and limitations of AVT, as outlined in

Table 5. Pros and cons of ambient vibration testing.

Advantages	Disadvantages
A convenient method for measuring structures under servicing loading. No special excitation devices are required. Non-destructive testing. The cost of the overall experiment is lower. Can be performed during the normal operation of structures. Practical for use in large and heavy structures.	Tedious setup, using hundreds of meters of electrical cables. Does not provide precise details about the exact location of the damage within a structure. Tough and shielded. Need to ensure minimal signal loss and interference over long distances. Estimated damping values are prone to errors. High-sensitivity sensors are required.

, underscore its practicality compared to FVT, particularly due to its feasibility during traffic operations, minimal requirement for costly excitation equipment, provision of reliable data, and lower total cost of experimental work. While FVT may offer more precise MI findings, the simplicity, cost-effectiveness, and practicality of AVT make it preferable for various scenarios, particularly for large and flexible viaducts where regulated excitation is challenging.

Note that the applicability of AVT is highly dependent on the characteristics of the structure itself and the environmental excitation conditions; thus, there are scenarios that may lead to its inapplicability or reduced reliability. For example, a lightweight structure where the vibration caused by environmental excitation is too small.

6. Conclusions

In this paper, we conduct a full-chain review of SHM, from data collection to data processing, modal identification, and finally structural damage detection. In our review, we first examine the wireless sensors and wireless networks suitable for SHM. We then investigate the existing modal identification methods. We further provide a review of structural damage detection methods, with a focus on state-of-the-art vibration-based methods. Finally, we review the pros and cons of the vibration-based methods for the SHM of viaducts.

Based on our review, we identify that the combination of long-range WSNs and vibration-based SHM is a promising method for attaining effective long-term viaduct structural monitoring. This combination addresses the inherent limitations of traditional wired SHM systems for urban viaducts, improving monitoring effectiveness and precision. Moreover, long-range wireless sensor-based SHM is highly beneficial for monitoring large viaducts or those located in remote areas.

Note that combining the wireless sensor approach with the vibration method provides non-destructive and cost-effective monitoring for the SHM of urban viaducts. Through the wireless examination of a structure’s dynamic response to excitations, these methods allow for the early detection of damage in the initial phase, enabling quicker responses and avoiding severe collapse. Overall, the integration of long-range WSNs with vibration-based methods is a cost- and energy-efficient solution for the long-term SHM of urban viaducts.

Structural health monitoring technology continues to progress. We envisage four key developments in the future: energy-harvesting sensor networks, enhanced SHM deployments, advanced machine learning-driven SHM data processing, and fusion of vision and wireless sensing-enabled SHM damage detection.