A Review of Vibration-Based Scour Diagnosis Methods for Bridge Foundation

Abstract

Foundation scour poses a serious threat to bridge safety in the whole life cycle and leads to many bridge failure incidents. Recently, as an important subfield of bridge structural health monitoring, vibration-based scour diagnosis methods have garnered widespread attention, particularly due to their rapid and low-cost features, which overcomes the difficulties of complex equipment installation associated with the traditional approaches. Recent advances of this method within the last decade are reviewed in this paper. Firstly, the principle of scour diagnosis and vibration excitation methods are introduced. Then, existing qualitative and quantitative studies on scour diagnosis are reviewed, respectively. The former refers to identifying the scour location based on the bridge dynamic characteristics or dynamic response changes, and the latter refers to identifying scour depth based on model updating or machine learning methods. Based on the above review, some important but neglected issues are summarized and discussed in depth, and some challenges and future trends are proposed, including innovative excitation methods, mitigation of environmental conditions interference, soil–structure interaction prediction and application of machine learning techniques.

1. Introduction

Foundation scour is one of the important reasons for bridge failure. In the United States, half of the more than 1000 bridge collapses from 1980 to 2012 were caused by scour [1,2]. The scour-induced average annual cost for highway repair has exceeded 30 million dollars in the 1990s [3], and the indirect economic loss caused by traffic interruption was even harder to estimate. In China, 58% of the 192 bridge failure cases from 1949 to 2021, as studied by Xiong et al. [4] were related to scour. For example, two piers of the Shi-Ting-Jiang Bridge collapsed due to severe scour during floods in August 2010 (Figure 1a), causing two carriages to fall into the river and disrupting the Baocheng Railway. Figure 1b shows the Min-Jiang Bridge destroyed due to severe scour during floods in July 2018, with three piers and 120 m long bridge deck collapsing in succession. Scour often leads to sudden bridge failures, which seriously threatens bridge safety in the whole life cycle and causes huge loss of life and economy [5,6,7].

Scour is a natural phenomenon in which the riverbed soil layer is eroded by the current [8]. The scour holes around piers are caused by general scour and local erosion. General scour is the result of the compression of the channel cross-sectional area by newly built piers. Local scour is caused by water flowing around the piers and abutments and contributes the most to scour hole [9]. Scour weakens the constraint effect of soil on foundation and reduces the stiffness and bearing capacity of foundation [10]. The bridge is prone to large displacement under the coupling effect of scour and flood or ship collision, leading to a significant collapse risk [11,12,13,14,15].

Scour depth prediction is the most direct method to reduce scour threat. Empirical formulas have been widely used in national design code [16,17], which refer to the relationship between the equilibrium scour depth and the scour-related parameters, such as river morphology, hydraulic properties, sediment properties and pier geometry, determined via field or laboratory tests [4]. Empirical formulas are simple to use but their prediction accuracy is low due to the complex scour-related factors [18]. Moreover, intelligence-based scour prediction methods, such as artificial neural network (ANN) [19,20], support vector machine (SVM) [21,22] and genetic algorithm (GA) [23], etc., provided more accurate results than empirical formulas [24]. However, both methods ignored influence of scour morphology on bridge safety [4]. The scour evolution in the field environment is an uncertain and hidden process, and it is very difficult to predict scour effectively in a long term.

Scour inspection is necessary during bridge operation, especially after severe flooding. Diving detection is the most commonly used method, but it is time-consuming, laborious and dangerous [25]. Some equipment used to detect scour condition, includes Remote Operated Vehicle, Time Domain Reflectometer, Image Sonar, Magnetic Rock, Ground Penetrating Radars, Fiber Bragg grating sensor, etc. [26,27,28,29,30,31,32,33], whose operational principles are stated in the literature [34]. However, most of these equipment requires professional personnel to install and operate underwater, which is inefficient and costly. Moreover, the detection result is also easy to be disturbed by the muddy water and loose accumulation at the scour hole [35].

Recently, the study on vibration-based bridge scour diagnosis has attracted widespread attention, which is an important subarea of bridge structural health monitoring [36]. Its basic principle is to detect scour indirectly based on the change of dynamic characteristics or dynamic response of bridges. Such method relies on the bridge structural health monitoring system or the temporary vibration data acquisition system, does not require expensive equipment and difficult underwater operations, so it is less disturbed by the underwater environment. Existing studies have shown that this method can achieve accurate, economical and rapid scour detection and structural integrity assessment [37,38,39,40,41,42,43]. Prendergast and Gavin [34], and Bao and Liu [37] reviewed the research progress of such methods prior to 2015, focusing on the structural frequency spectrum and the advances in scour monitoring sensors. Recently, many new methods were proposed to explore new directions. Kazemian et al. [38] published a review in 2023, focusing on several traditional scour monitoring techniques and scour monitoring based on structural dynamic characteristics. Although these reviews are significant, a systematic and hierarchical review for recent progress and new trends is still lacking, especially for important studies, such as dynamic-response-based scour diagnosis and scour depth identification.

Accordingly, this paper provide a comprehensive review of the latest progress of vibration-based scour damage in the following order. Firstly, the principle of scour diagnosis and vibration excitation method are introduced in Section 2. Then, existing qualitative and quantitative studies on scour diagnosis are reviewed in Section 3 and Section 4, respectively. The former refers to identifying the scour location based on the bridge dynamic characteristics or dynamic response changes, and the latter refers to identifying scour depth based on model updating or machine learning methods. Based on the above review, some important but neglected issues are summarized and discussed, and challenges and future trends are proposed in Section 5. The purpose of this work is to provide some inspiration for future research on scour diagnosis. The overall review process of this paper is shown in Figure 2.

2. The Principles and Excitation Method of Vibration-Based Bridge Scour Diagnosis

The principles of vibration-based scour diagnosis are introduced in Section 2.1 to provide the basis for the subsequent review. Common vibration excitation methods are introduced in Section 2.2, including forced excitation, ambient excitation and moving vehicle excitation.

2.1. The Principles of Vibration-Based Bridge Scour Diagnosis

Bridge scour changes the structural dynamic characteristics and affects the structural dynamic response. For structural damage diagnosis, the most direct way is to judge the structural stiffness loss by comparing the expected frequency with the measured frequency [37]. Stiffness and frequency of piers have the following general relationship [44]:

$$f_n=\frac{1}{2\pi }\sqrt{\frac{k}{m}}$$

(1)

where, f_n is pier frequency, and k and m are stiffness and mass, respectively.

Equation (1) shows that f depends on k and m. Piers are generally simplified to cantilever beams with concentrated mass and lateral constraints at their ends. Scour removes the soil around the foundation, resulting in the increase in the free length of piers and the decrease in the horizontal stiffness k. The slight influence of pier damage on its mass m can be ignored. Therefore, if the measured pier frequency is significantly lower than the expected frequency, it is reasonable to infer that foundation scour has occurred. However, a non-negligible problem is that the superstructure damage, such as bridge bearing emptying, reduce the lateral constraint at the pier top, leading to the decrease in f. Fortunately, bearing damage is easily found in daily inspection, so its effects can be appropriately ignored [37]. In other words, since both scour and bearing emptying reduce pier frequency, they can be considered together in the initial diagnosis and distinguished in the refined detection [45]. Similarly, scour also affects the dynamic characteristics, such as mode shape and flexibility, which are not discussed again.

The relationship between the time-domain and frequency-domain responses of the damaged structure is as follows:

$$x(t)=\frac{1}{2\pi }{\displaystyle {\int }_{-\infty }^{+\infty }X(\omega )e^{-i\omega t}d\omega }$$

(2)

$$X(\omega )=H(\omega )·F(\omega )$$

(3)

where, x(t) is the structural dynamic response, X(ω) is the response spectrum, H(ω) is the frequency response function and F(ω) is the external excitation spectrum.

Equation (2) shows that scour affects the structural dynamic response, so the difference comparison between the measured and the expected response is regarded as another approach for scour detection. Equation (3) shows that x(t) depend on H(ω) and F(ω). H(ω) depends on the structural state. When F(ω) is constant, the structural response depends only on the structural state, otherwise the relationship will be complicated. Therefore, the key of scour diagnosis based on dynamic response is to keep the frequency spectrum of external excitation constant, or to adopt indicators insensitive to the external excitation change. Fortunately, the common excitation methods, such as pulse excitation and ambient excitation, meet the requirements of spectrum stability to some extent.

2.2. Excitation Method of Vibration-Based Bridge Scour Diagnosis

2.2.1. Forced Excitation

Forced excitation is the dynamic loads imposed by hammers, shakers, etc., whose amplitude and frequency can be predetermined to generate the desired vibration signals with high signal-to-noise ratio [37,46]. These advantages are beneficial to identify structural dynamic characteristics more accurately [47,48]. In the study on bridge scour diagnosis, Nishimura [49] used the swinging impact force of a heavy hammer to vibrate pier, and first proposed the impact vibration test method (IVTM) to test pier frequency. Zhan et al. [41,50,51] used IVTM to conduct a large number of experimental studies on bridge scour diagnosis. Other studies have used IVTM to test frequencies of scaled pier models at different scour stages, with impact forces applied using hammers, iron balls, etc., and experimental results showed that the frequency decreased with scour development [47,48,52]. Elsaid and Seracino [10] used an impact hammer to excite a steel plane frame model to study the sensitivity of the flexibility and curvature of the superstructure to scour. In conclusion, forced excitation has been widely used in scour diagnosis of laboratory scaled models and medium-small span bridges. It should be noted that excessive impact force may cause structural damage. Moreover, in the field test, the traffic needs to be closed and the vibration exciter needs to be installed on site, which is costly and time-consuming [37].

2.2.2. Ambient Excitation

Ambient excitation refers to wind, flood, earthquake and other unintentional natural forces [37]. Unlike forced excitation, ambient excitation is random and unmeasurable, and is usually assumed to be white noise. Vibration testing under ambient excitation does not need to close the traffic and install vibration exciter, which is safer and more convenient. However, due to the strong randomness of vibration and interference from the environment [53,54], the process of identifying structural parameters based on ambient vibration and evaluating structural state is generally complicated. Samizo et al. [47] identified the pier frequency based on the pier micro-vibration during flood, and found that the frequency at high water level was closer to the test result of IVTM. Liu et al. [55] measured the transverse frequency of the scoured pier of a steel truss bridge using the natural pulsation method, and further evaluated the safety condition of the pier using the fitness index. Ambient excitation has significant effects on large structures and was more commonly used in the scour diagnosis for long-span bridges. Xiong et al. [43,45,56,57], Chen et al. [58] and Li et al. [59] studied scour diagnosis methods for cable-stayed bridges based on ambient vibration, which will be introduced in detail below.

2.2.3. Moving Vehicle Excitation

During bridge dynamic load testing, moving vehicle excitation has the attribute of forced excitation, and its effect depends on the vehicle weight, speed and road conditions. In bridge long-term health monitoring, random traffic load belongs to ambient excitation. Under moving vehicle excitation, the dynamic characteristics of the bridge are identified using the output-only method or the multi-input multi-output method [60,61,62], the latter of which is used in mobile impact vibration testing [63,64,65]. In addition, Yang et al. [66,67] proposed testing bridge frequency using vehicle response, and relevant research is being further promoted [68].

In the study on scour diagnosis, moving vehicles are widely used to excite horizontal vibration of bridges, and then the frequency sensitive to scour is extracted using the recorded vibration signals. For example, Ko et al. [46] used the transverse velocity response of bridge deck to extract the transverse bridge frequency, and Prendergast et al. [69,70,71] used the longitudinal acceleration response to extract the longitudinal bridge frequency. The research showed that the moving vehicle was an ideal excitation method for medium-small span bridges due to simple operation, strong mobility and low cost. However, the horizontal vibration of the bridge caused by normal moving vehicles was significantly weaker than the vertical vibration, and the horizontal frequency sensitive to scour cannot be fully excited, which is unfavorable to vibration testing and scour diagnosis. Focusing on this problem, Li et al. [72] studied the effect of vehicle braking on bridges, and found that braking action stimulated significant structural longitudinal vibration sensitive to scour. Further, scour diagnosis methods based on braking excitation were proposed and validated via numerical simulation [73,74]. This research developed a new application scenario for moving vehicle excitation.

3. Vibration-Based Qualitative Diagnosis of Bridge Foundation Scour

The purpose of the qualitative diagnosis of scour is to determine whether scour occurs and find the location of scour; this can be realized using methods based on structural dynamic characteristics or dynamic response [39]. The research progress of these two methods in the last decade is reviewed in Section 3.1 and Section 3.2, respectively.

3.1. Bridge Scour Diagnosis Based on Dynamic Characteristics

3.1.1. Frequency-Based Scour Diagnosis

Frequency-based scour diagnosis methods have received much attention due to the easy access to frequencies of the bridge. Existing studies are introduced below in chronological sequence, including numerical simulation, laboratory tests and field tests.

The variation of pier frequency with scour development has been investigated using laboratory tests [52,75]. The test was performed in a large flume (Figure 3a). A concrete column with 4 m height and 0.45 m diameter was buried in compacted sand for 0.3 m to simulate a pier with a shallow foundation, and two concrete bridge decks were supported on the pier top. A motion sensor was installed near the pier top to collect the acceleration response. The vibration of the pier before and after water injection in the flume was generated by rubber hammer and flow, respectively. In the test, the flow velocity in the flume gradually increased from 0 to 0.6 m/s, resulting in the development of scour hole. Briaud et al. [75] modified the above model by installing 8 steel bars at the bottom of the concrete column to simulate the pile foundation, and conducted the test in the same process. The collected acceleration data was transformed using Fast Fourier Transform (FFT) to extract the pier frequency. The results of the two tests show that the first three order frequencies in the flow direction decrease significantly with the scour development (Figure 3b), while the frequency in the traffic direction remains unchanged. In addition, Yao et al. [76] also conducted field tests on two bridges in Texas. The accelerometer was installed on cap beam to collect vibration under traffic load. However, the reliable pier frequency was not obtained due to the discontinuous measured acceleration data, indicating that it may be difficult to test the pier frequency with noise interference.

Ko et al. [46] proposed using the frequency spectrum of bridge transverse response to diagnose scour. Firstly, the influence of scour on the dynamic characteristics of a simply supported bridge with caisson foundation was studied via numerical simulation. The results showed that the horizontal longitudinal (HL) and horizontal transverse (HT) frequencies of the bridge decreased with scour development, and the latter decreased more, indicating that the HL and HT frequencies can be used to diagnose scour. After that, the vibration data of the two simply supported bridges under traffic load was recorded with velocity sensors. Among them, the scour depth of two piers of Bridge I with caisson foundation is 0.5 m and 6.5 m, respectively, and the sensors were installed on the deck directly above the two piers. The scour depth of one pier of Bridge II with pile group foundation changed from 4.5 m to 7.5 m, and the sensor was installed on the cap beam. The average Fourier spectrum of the velocity response was taken as the scour indicator (SI). SI in HL and HT directions of Bridge II was shown in Figure 4, indicating that the HT frequency changes significantly due to scour, while the HL frequency changes little due to deck constraints. This is reasonable because scour mainly affects the pier frequency in the flow direction.

Prendergast et al. [77] studied the scour effect on the frequency of the pile through experiments. Scale model tests were carried out in a square steel box filled with compacted sand. A 1.26 m long rectangular steel pipe was partially buried in sand to simulate pile, and an accelerometer was installed on the pile top. Five scour levels were simulated by gradually removing the soil layer in 5 cm scour increments. Pile vibration was excited using an impulse force from a swinging arc mechanism, and the pile frequency was obtained by processing the measured acceleration data using FFT. Experiments were performed on four steel piles with different section stiffness, and two cases with and without water were designed to study the influence of water. The results showed that the pile frequency decreased with scour development. The water effect on the frequency of flexible pile was obviously greater than that of the rigid pile, and the frequency of rigid pile in water was close to that in air. In addition, a field experiment was designed to investigate frequency-based scour prediction. A large steel pipe of 8.76 m in length and 0.34 m in diameter was partially buried in sand to simulate the cantilever pile. The pile was excited using a pulse hammer and its frequency was extracted using FFT. Moreover, a numerical model was developed to predict the pile frequency at different scour levels. By comparing the prediction results with different soil-spring stiffness calculation methods, it was found that the predicted pile frequency based on small strain stiffness method was closer to the measured frequency (Figure 5). The results showed that it was feasible to predict scour depth based on frequency.

Thereafter, Prendergast et al. [40,69,70,71] conducted a series of research on frequency-based scour diagnosis of pile foundation. A vehicle–bridge–soil dynamic interaction (VBSD) model was developed to calculate the structural dynamic characteristics and vehicle-induced response. The scour effect on the frequency of a two-span integrated bridge was studied by removing the constraint of piles above the scour line. The results showed that the bridge fundamental frequency (longitudinal global sway) decreased significantly under severe scour condition (10 m), and that the frequency can be identified by using FFT to process the vehicle-induced response of bridge [71]. Three soils with different stiffness were considered, including loose, medium and dense sand, and it was found that frequency was not sensitive to soil stiffness. Further studies showed that [40] the local vibration modes of bridge piers were only affected by scour at nearby locations, indicating that the scour location can be directly determined based on the change of pier frequency.

The above studies independently considered the dynamic action of water or vehicles, and Kong and Cai [78] considered the coupling effects of both in bridge scour diagnosis. A vehicle–bridge–wave interaction model was developed based on the time-varying wave force calculated using the Morison equation. The scour effect on dynamic response of the bridge under wave force was studied, and the numerical results showed that the wave force was an effective lateral excitation, and the lateral displacement and acceleration response at pier top excited by wave force changed obviously due to scour. Scour-induced frequency change of the pier can be obtained by processing the lateral response of the pier with FFT. Moreover, scour effect on vehicle response in vehicle–bridge–wave interaction system was studied by comparing the vertical acceleration responses of two independent trailers in a passing vehicle at different scour levels. It was found that the vehicle response changed obviously due to scour (Figure 6a), and the bridge frequency sensitive to scour was obtained by processing vehicle residual response using FFT (Figure 6b). The results showed that vehicle response can be regarded as a potential scour indicator.

Bao et al. [79] studied two critical problems in scour diagnosis using model tests, namely, the influence of sensor location and scour hole shape on pier frequency. Four scaled pier models (Figure 7a) were partially embedded in a plastic tank filled with compacted sand. Pier vibration was generated by modal hammer hitting, and pier frequencies were obtained by processing vibration data collected via accelerometers using FFT. The frequency change of different piers with scour development was first tested, and the results (Figure 7b) showed that greater the pier stiffness, the faster the decrease in frequency. In the test, the accelerometers were installed at different positions on the pier surface, and the results showed that the identified frequency was the same at the same horizontal position, but the identified frequency at the pier bottom was greater than that at the pier top due to the constraint effect from the soil. Moreover, a criterion was proposed to define the depth of asymmetric scour holes using the average upstream and downstream scour depth. Experimental results in Figure 7c showed that this criterion can accurately quantify the effect of asymmetric scour on frequency.

Based on the above laboratory tests, Bao et al. [80] also studied the influence of soil property on pier frequency. The piers shown in Figure 7a were sequentially buried in sand and clay, and their frequencies were measured at different scour levels. The experimental results showed that the pier frequency in sand was greater than that in clay at the same scour level due to the larger elastic modulus of sand. Soil elastic modulus was the key factor affecting pier frequency rather than the soil type. Moreover, the influence of pier diameter on pier frequency prediction based on the Winkle model was studied. Lateral soil spring stiffness of pile was calculated using the American Petroleum Institute Method (APIM) and the small strain stiffness method (SSSM). By comparing the measured frequency of piers with the predicted frequency, it can be found that the influence of pier diameter does not need to be considered when using APIM for sand. When using SSSM for sand or clay, the pier diameter effect should be considered only if the elastic modulus of the soil varies with depth.

Although the above laboratory tests focused on the effect of soil, the scaled soil model could not accurately simulate the property of the full-scale soil. To overcome this problem, Kariyawasam et al. [81] innovatively simulated the stress distribution of full-scale soil in the scaled laboratory model using centrifugal test. Three piers with single pile, pile bent and shallow foundation were designed. Pier vibration was excited using an automatic modal hammer, and the accelerometer was installed on the pier top to collect vibration data for identifying pier frequency. Scour was simulated by removing the soil around the foundation. The experimental results showed that the frequency of pile bent, single pile and shallow foundation models decreased by 44%, 23% and 4%, respectively, when foundation embedment loss reached 30%. The results showed that frequency-based scour diagnosis was applicable to pile foundation rather than shallow foundation.

In addition to the above theoretical and laboratory studies, some recent studies have applied frequency-based scour diagnosis methods to practical engineering, developing specialized scour detection platforms or utilizing bridge structural health monitoring system (BSHM) to monitor scour.

Cheng et al. [42] developed a mobile vibration measurement and signal analysis platform for rapid scour detection. The platform, which consists of signal acquisition, reception and transmission equipment and a self-designed signal analysis system, was integrated on a small trailer. During scour detection, this platform was driven to the bridge and kept stationary. The accelerometer installed on the platform was used to collect traffic-induced vibration signals, and the bridge frequency in the flow direction, which is sensitive to scour, was obtained by analyzing the collected data using FFT. Field tests on several bridges in Taiwan showed that the platform has the advantages of simple operation, strong maneuverability and high accuracy. Combined with the existing studies [82], the critical frequency ratio Rc (i.e., transverse frequency ratio of scoured and healthy bridge) was proposed as the scour indicator. Rc decreased with scour development and can be calculated using the detection results of the mobile platform at different times.

Xiong and Cai [45] proposed a scour diagnosis method for long-span bridges based on the trend-change detection of bridge frequency. First, Empirical Mode Decomposition and Short-time Fourier Transform were used to process the vibration monitoring data from the BSHM system to extract the instantaneous frequency of the bridge. Then, a normalized probability distribution model of frequency was obtained based on kernel density estimation. Finally, a control chart method was used to detect frequency anomalies and determine the scour condition at a specific significance level α. The method was numerically verified by a case study of a two-tower cable-stayed bridge, the results showed that the trend of instantaneous frequency of the bridge changed significantly with scour development, which was an effective scour indicator. The larger the α, the higher the sensitivity of scour diagnosis, but the lower the confidence level. In addition, a field case was provided. By analyzing the measured data from the BSHM system of Anqing Yangtze River Cable-Stayed Bridge, it was found that the bridge frequency trend was consistent with the seasonal scour condition, which proved the potential of this method.

In conclusion, the horizontal frequency of pier and bridge deck can be used as potential indicators in the scour diagnosis. Pier frequency is more sensitive to scour because scour directly weakens the lateral constraints of the pier, resulting in significant changes in the local frequency of the pier. The frequency of the bridge deck is more convenient to measure due to the ease of sensor installation, but it is less sensitive to scour because scour affects deck indirectly by changing the stiffness of the substructure. The scour-frequency relationship of the long-span bridge is complicated due to its complex structural system, but the frequency-based scour diagnosis is also applicable. At present, relevant research mainly focus on numerical simulation and laboratory test. Considering measurement error, noise interference and difficult to excite the horizontal modes of real bridge, more attention should be paid to field tests in the future.

3.1.2. Mode Shape-Based Scour Diagnosis

The measurement accuracy of mode shape is lower than that of frequency but it is more sensitive to local stiffness loss. Measurement of vibration mode requires scattered installation of sensors, which is difficult for piers, so it is common to test the mode shape of the superstructure to establish scour indicators. The relevant progress on different bridge types is reported below.

For continuous girder bridges, Elsaid et al. [10] proposed a scour diagnosis method based on the mode shape of the superstructure. A two-span continuous bridge was simulated using a steel grid supported on three rows of steel piles with variable length. The change of pile free length caused by scour was simulated by adjusting pile length. The model vibration was excited using a modal hammer, and the horizontal and vertical mode shapes of the bridge deck for various pile lengths were measured. The test results showed that with the pile length gradually increasing from 1.1 m to 1.7 m (equivalent scour depth was 0.6 m), the horizontal mode shape changed obviously, while the vertical mode shape remained basically unchanged. In order to improve the accuracy of scour diagnosis, two indicators of flexibility-based deflections and flexibility-based curvature were calculated based on the horizontal mode shape of superstructure, and the test results showed that the former was more effective in scour diagnosis (Figure 8).

For long-span cable-stayed bridges, Xiong et al. [57] proposed four scour diagnostic indicators, including frequency change rate (FCR), modal assurance criterion (MAC), flexibility-based deflection (G) and mode shape curvature (Φ), and verified and compared their effectiveness through numerical simulation. The modes of a single-tower cable-stayed bridge were analyzed using the finite element method. The pile–soil interaction was simulated using the Winkler model and API method, and the scour was simulated by removing the soil lateral springs on the pile. The simulation results showed that the low order horizontal and torsional modes of the bridge deck were sensitive to scour. The mapping relationship between the four indicators and the scour depth was calculated based on these sensitive modes. It was found that MAC was an invalid indicator with low sensitivity; FCR, Φ and G could infer whether scour occurred; and G could determine the scour location. This study provided a potential method for the scour diagnosis of long-span bridges.

For multi-span simply supported girder bridges with shallow foundations, scour may weaken the vertical stiffness of foundation and then change the vertical mode shape of bridge decks. Accordingly, Malekjafarian et al. [83] proposed a scour diagnosis method based on mean normalized mode shape (MNMS), which was verified via numerical simulation and laboratory tests. A numerical model of a 6 × 20 m simply supported bridge (Figure 9a) was constructed, and scour was simulated by reducing the foundation stiffness K_f. The modal analysis results show that the first-order mode shape of bridge deck (vertical synchronous vibration) was sensitive to scour, so the normalized amplitude of this mode at pier position was defined as the scour indicator, called MNMS. MNMS at scour locations increased with decreasing K_f, as shown in Figure 9b. Moreover, laboratory tests were performed using a scaled four-span simply supported bridge (Figure 9c) design based on the numerical model. The foundation stiffness reduced by 25% or 45% when the foundation spring was replaced with a smaller stiffness. Bridge mode shape was identified under moving vehicle excitation using the frequency-domain decomposition method. The experimental results showed that MNMS can accurately identify scour at single and multiple locations. MNMS are insensitive to random noise and vehicle parameter changes, and have good robustness.

Khan et al. [84] improved the method proposed by Malekjafarian et al. [83]. In order to reduce the number of sensors required in the modal test of multi-span bridges, the distributed modal analysis method was used. In one test, only two accelerometers installed at the midspan and supports were used to measure the half-span mode shape of a multi-span bridge. As the accelerometers were moved to next half-span, next tests were performed. Finally, all mode shape segments were spliced to obtain the global mode shape. A laboratory test was conducted to verify the method using the model in Figure 9c. The test results showed that the mode shape measured using the decentralized method was basically consistent with the traditional method. Further, two scour levels with foundation stiffness loss of 25% and 45% were simulated using the model, and a mode shape curve fitted via the maximum likelihood method was used as a scour indicator (SI). Experimental results showed that scour location can be determined based on the SI changes. Moreover, with the increase in scour level, the root-mean-square difference of SI between health and scour conditions gradually increased.

In conclusion, mode shape-based scour diagnosis is mainly used for multi-span bridges. Scour indirectly affects the mode shape of the superstructure by weakening the foundation stiffness. Scour mainly weakens horizontal stiffness for pile foundation and vertical stiffness for shallow foundation. Therefore, for bridges with these two types of foundations, horizontal mode shapes and vertical shapes are selected, respectively, to construct the scour indicator. The feasibility of these methods has been verified via some laboratory tests, but the simulation of scour is relatively simple and idealized, and the designed severe scour condition may not represent the real scour state. Hence, more field studies should be carried out in the future.

3.2. Bridge Scour Diagnosis Based on Dynamic Response

Scour diagnosis based on dynamic response is another potential direction. This method can fully retain and directly utilize the time information in the vibration signal. Vibration signals are analyzed directly using time-domain or time–frequency domain analysis methods, such as wavelet transform, and scour indicators are extracted combined with statistical analysis. At present, relevant research is mainly focused on middle-small span bridges, and structural vibrations are excited by moving vehicles. Section 2.1 points out that excitation method is a critical factor in dynamic response-based scour diagnosis. Therefore, research progress is introduced according to the different excitation effects of vehicles on bridges.

3.2.1. Scour Diagnosis Based on Dynamic Response Excited by Normal Traffic Loads

Normal traffic load can stimulate the significant vertical dynamic response of bridge deck. For vehicle–bridge coupled vibration systems, both bridge and vehicle dynamic responses can be used to identify scour.

Foti and Sabia [39] proposed utilizing dynamic characteristics and vehicle-induced response of bridge to detect scour. The case study is a simple-supported bridge with 15 m long pile foundation. The flood caused a 6 m deep scour hole around one of the bridge piers. After that, the bridge was repaired by replacing the scoured piers. Field dynamic tests were performed before and after the bridge repairing, and the acceleration response on the bridge deck and foundation mat was recorded when heavy vehicles passed. The frequency and mode shape of the bridge deck were identified using the collected response, as shown in Figure 10a, where the scour hole was located at 60 m. Figure 10a shows that there are differences in frequency and mode shape between the second span and other spans before repairing, and these differences basically disappeared after repairing, indicating that the change in the dynamic characteristics of the bridge deck is caused by scour, not damage in the span. In addition, given that the scour holes were non-uniform on the upstream and downstream sides, the covariance of acceleration response on the foundation mat was selected as the scour indicator to monitor scour. The collected responses were analyzed, and the results showed that there was a significant difference between the indicators of scoured pier and healthy pier before repairing, and this difference disappeared after repairing (Figure 10b). Moreover, when vehicles pass the bridge deck from the upstream and downstream sides, respectively, the indicators of the scoured piers were obviously different while the indicators of healthy pier were basically the same. The results show that both modal identification for bridge spans and dynamic response analysis of piers are potential tools for scour monitoring.

For multi-span simply supported bridges with shallow foundations, OBrien et al. [85] proposed using wavelet-based operating deflection shapes (ODS) to identify scour locations. A four-span simply supported bridge model (Figure 9a) and a half-car model were used for numerical studies. The vertical acceleration response of bridge pier excited by moving vehicle was translated into the space domain to match the driving path. The tests were repeated using a random fleet, and the acceleration signals collected in multiple tests were successively averaged and wavelet transformed. Finally, the self-power spectrum of wavelet coefficients was used to calculate ODS. The difference of ODS between healthy and scour conditions was defined as a scour indicator. The results of numerical cases showed that the indicator can identify single or multiple scour locations when the foundation stiffness loss was greater than 25%, but the healthy locations may be mistaken for scour locations.

Several recent studies have attempted to diagnose scour using response of passing train or automobile combined with statistical analysis. Fitzgerald et al. [86] utilized train response to monitor bridge scour and conducted a numerical study. The train–bridge coupled vibration response was calculated using the bridge numerical model in Figure 9c (extended to eight spans) and a quarter-train model. The average wavelet coefficients of vertical acceleration of train bogie were obtained using continuous wavelet transform for all simulation results of a group of trains. The average wavelet coefficient difference between health and scour condition was defined as a scour indicator. The numerical results showed that scour locations can be identified even without any prior knowledge, and the identified accuracy can be improved by increasing the number of trains. This method cannot identify scour depth but can be used for preliminary detection.

Zhang et al. [87] proposed using vehicle response and probability statistics methods to detect scour, and conducted numerical and experimental studies. The vehicle–bridge coupled vibration response was calculated using the bridge model (reduced to four spans) in Figure 9a and a quarter-vehicle model. The wavelet packet energy E of vehicle response was calculated, and the normal probability distribution of E (expressed as N(μ, σ)) was fitted based on the simulation results of a group of random vehicles. The numerical study showed that the probability distribution was sensitive to scour, and scour resulted in a significant increase in the mean value μ and a positive shift in probability distribution. Further, in order to identify scour location, the vehicle response was divided equally into each span to match the driving path, and the relative change indicator ID of the mean value μ of each signal segment was defined as the scour indicator. Numerical studies showed that ID can accurately identify scour location. Laboratory tests were conducted to verify the above conclusions using the model in Figure 9c. The test results also showed that using vehicle axle acceleration leads to better diagnostic results than vehicle body acceleration.

In conclusion, the dynamic response-based scour diagnosis method directly processes the response signal to extract the scour indicator, eliminating the modal identification. Experimental parameters, such as vehicle parameters and road roughness, are generally random, resulting in random excitation input and structural response output. Therefore, one difficulty of such method is to overcome the influence of the changes in experimental parameters through statistical analysis for a large number of response data. Both bridge response and vehicle response were used in diagnosis. The effectiveness of the former has been verified using numerical and field experiments; the latter is an indirect drive-by detection, which can significantly improve the detection efficiency. Only a few numerical and laboratory tests have been carried out for specific bridge, and its feasibility still needs to be further verified.

3.2.2. Scour Diagnosis Based on Dynamic Response Excited by Vehicle Braking

Existing studies have shown that the horizontal vibration of the bridge is sensitive to scour. Normal passing vehicles mainly stimulate the vertical vibration of the bridge, while significant horizontal vibration was difficult to generate under traffic load [46,78], which is not conducive to the field scour diagnosis. Recently, in order to solve the above problem, the vehicle brake excitation was used to stimulate the large horizontal vibration of the bridge. Some numerical simulation work has been performed and relevant progress is described below.

A vehicle–bridge coupled vibration model considering vehicle braking was established [73]. The braking force coefficient (the ratio of braking force to vehicle weight) was assumed to be a ramp function. The time-varying braking force of the axle was derived using the condition of axle deformation and force balance during braking. Finally, the dynamic response of the vehicle–bridge system was calculated using the modal superposition technique. The model has been validated through field tests and was the basis of the subsequent numerical research.

Using the above model, Li et al. [72] compared the influence of foundation scour on the dynamic response of bridges under two excitation modes of vehicle normal running and braking. A four-span concrete continuous beam bridge with pile foundation and a 29.8 t three-axle truck was used for the numerical study. Multiple scour levels were simulated. The numerical results showed that under these two excitation modes, the longitudinal dynamic response of the pier top was sensitive to scour, and the peaks of longitudinal displacement and acceleration increased significantly with the increase in scour depth (Figure 11). However, the response amplitude under vehicle braking excitation was significantly higher than that under normal running excitation, and its spectrum mainly contains low frequency components. It can be concluded that vehicle braking may be a better excitation method than normal traffic load in scour diagnosis.

Recently, several scour diagnosis methods have been proposed based on vehicle brake excitation and verified via numerical simulation. Zhang et al. [73] applied dynamic response cross−correlation index to conduct scour diagnosis. The longitudinal acceleration response of the bridge excited by vehicle braking was extracted to calculate the cross-correlation index, CorV. Theoretical derivation and numerical simulation showed that CorV had fixed shape characteristics, and its shape depended on the structural state. Importantly, CorV was not affected by the changes in test parameters, such as road roughness and braking-related parameters. Therefore, a scour indicator D_CorV was proposed based on the relative variation of CorV between intact and scour states. The validity of scour indicator was verified though a numerical case. The results showed that the amplitude of D_CorV increased with the increase in scour depth, and D_CorV could accurately identify scour at single or multiple locations (Figure 12). The proposed method can be used for the rapid scour detection of highway girder bridges with pile foundation.

Yang et al. [74] proposed a scour diagnosis method for highway continuous girder bridge using wavelet packet analysis. The longitudinal acceleration response of the bridge excited by vehicle braking was extracted to calculate the wavelet packet energy. The wavelet packet energy variance variation rate (WPEVVR) was defined as a scour indicator to identify scour location. A numerical example was used to verify the method, and the results showed that WPEVVR can accurately identify scour at single or multiple locations. Further, a scour depth inversion method was proposed. A sufficient number of representative scour cases were obtained through numerical simulation, and the functional relationship between WPEVVR and scour depth was fitted based on these cases. Finally, the WPEVVR tested in the field was substituted into the relationship to determine scour depth. The numerical study showed that this method can identify the scour depth accurately.

In conclusion, the scour diagnosis method based on vehicle brake excitation has the advantages of high precision, good anti-noise and easy usability. It can be used for rapid detection of foundation scour after flood, and can also be integrated into regular bridge dynamic test, which has good potential. In the future, such methods need to be further validated and improved through laboratory and field experiments.

4. Quantitative Diagnosis of Bridge Foundation Scour Based on Model Updating or Machine Learning

In damage diagnosis, damage quantification is a higher level and more challenging work than damage localization [88]. Scour depth quantification mainly uses model updating technology and machine learning technology. The former is a widely used traditional method, and its research progress is introduced in Section 4.1. As a new technology, the latter is currently receiving increasing attention and is introduced in Section 4.2.

4.1. Scour Depth Identification Based on Model Updating

4.1.1. Scour Depth Identification for Long Span Bridges

The basic idea of finite element model updating (FEMU) is to optimize the parameters of the finite element model (FEM), make the theoretical results coincide with the experimental data, and finally obtain a FEM that can accurately simulate the actual structural state [89]. Based on the FEN type, this method can be divided into direct model updating and proxy model updating surrogate models [90]. At present, FEMU has been widely used to identify scour depth of bridge foundation, and its research progress on long-span bridges is detailed below.

Chen et al. [58] identified the foundation scour depth of Gaopingxi cable-stayed bridge through model updating. The initial FEM was constructed using the bridge design parameters, and soil–structure interactions were simulated using the Winkler model. The frequencies of the girder and piers were identified from the ambient vibration of the bridge and were used to correct the boundary support conditions of the girder. Then, soil spring stiffness was corrected using the known deposit height of the pylon and the girder frequencies sensitive to scour. The soil spring stiffness at the pylon and pier locations was considered to be the identical in the same area. Finally, the scour depth at the pier was determined by fitting the measured pier frequency and by correcting the constraint height of the soil spring. The identified scour depth was validated via a field measurement. In the above process, the known soil deposit level at pylon was a critical condition for successful identification of scour depth at pier because it reduced the complexity of correcting soil spring stiffness. However, other bridges may not be able to provide this condition if the pylons and piers were located in the channel and cannot be observed.

Xiong et al. [43] proposed a two-stage scour depth identification method for cable-stayed bridges. In the first stage, the lateral soil spring stiffness of pile foundation was updated using the frequencies of the superstructure insensitive to scour. In the second stage, the scour depth of pile foundation was updated using the frequencies of superstructure sensitive to scour. Compared with the method used by Chen et al. [58], the proposed method does not require a reference pier with known soil deposit level and has fewer restrictions on bridge field conditions, so it is suitable for general cable-stayed bridges. The proposed method has been applied to Hangzhou Bay Cable-stayed Bridge. Eleven different frequencies of the bridge were identified from the ambient vibrations monitored on site. Using the initial FEM of the bridge and the above two-stage process, the scour depth of the bridge was updated with the integrated frequency difference F_d as the target parameter. The accuracy of the method has been proven by the field underwater measurement.

The environmental condition has a significant influence on the dynamic characteristics of long-span bridges and thus interferes with scour identification. In order to identify scour depth more accurately, Li et al. [59] proposed a method to filter out the influence of environmental conditions from the time-varying modal parameters. Hangzhou Bay Cable-stayed Bridge was selected as a case study. The FEM of the bridge was constructed to determine the frequency sensitive to scour through a modal analysis. Before opening to traffic, the natural frequencies of the bridge were extracted from monitored ambient vibrations using the frequency-domain decomposition method. The objective function was established to update the initial FEM using the transverse modes of the pylon and vertical modes of the girder, and the mapping relationship between the critical frequencies and scour depth was further determined. After opening traffic, the influence of environmental conditions on the measured frequency was filtered out using the nonlinear principal component analysis. Finally, the scour depth can be determined based on the reconstructed frequency and the mapping relationship. The accuracy of the proposed method was verified by comparing the results of diver’s visual detection.

4.1.2. Scour Depth Identification for Middle-Small Span Bridges

Compared with long-span bridges, middle-small span bridges with low structural redundancy and safety reserves are more likely to be damaged by flood and scour. Recently, scour depth identification for such bridge has attracted much attention.

Zhan et al. [41] proposed a scour depth identification method for highway bridges with pile foundation based on pier frequency and model updating. The working process is as follows. The modal hammer was used to apply longitudinal pulse to the pier. The Frequency Response Function (FRF) was calculated using the recorded force signal and the longitudinal acceleration signal of the pier to identify the pier frequency, then the objective function of model updating was constructed using the FRF correlation index and pier frequency. The damage index of structure was iteratively optimized to make frequency calculated by the FEM match the measured frequency, and the scour depth can be determined from the final damage index. The effectiveness of the proposed method was proven via numerical simulation and field test. It is worth noting that, similar to the method proposed by Chen et al. [58], since both foundation scour and soil properties affect pier frequency, it is necessary to conduct tests on other observable piers in the same area to obtain the soil resistance coefficient m, which is used to construct FEM of unobservable piers before model updating. Although the proposed method requires the installation of a working platform near the pier, it is safer and more convenient than underwater equipment.

In order to identify the scour depth of continuous girder bridges, Liao et al. [25] used an improved genetic algorithm (GA) to update the FE model. The characteristics of the improved GA are that the initial population obeys exponential distribution and uses gradient-like computation as the mutation operation. The proposed method was verified via numerical simulation. The dynamic response of a three-span continuous bridge was calculated by using the mass-spring model to simulate passing vehicles. The free vibration signal of the bridge was extracted to identify the bridge frequency. The free vibration signals of the bridge were extracted to identify the transverse bridge frequencies sensitive to scour using the SSI algorithm. Finally, an objective function was constructed using critical frequency and modal assurance criteria, and the model was updated using the improved GA. The numerical results showed that the proposed method can accurately identify the scour depth, with an error of less than 0.31 m.

He et al. [91] identified the scour depth of continuous girder bridges using the Kriging-assisted model updating method. The frequency difference and MAC of the bridge transverse mode were used to construct the objective function, and the non-dominated sorting GA algorithm was used to optimize the function. Numerical results show that this method can significantly shorten the model updating time. Only one sensor on each pier is needed to achieve satisfactory results.

Chen et al. [92] proposed a scour monitoring technology for offshore wind turbines with pile foundation based on Cross Model Cross Mode Method (CMCM). The basic idea was to construct the CMCM equation based on the measured and theoretical modes of the structure, and determine the scour depth by updating the stiffness correction coefficient of each element. The method was verified via a laboratory test. The FE model corresponding to the scaled single pile model was established using the concentrated mass method. The soil–structure interaction was simulated by the Winkler model and the equivalent soil spring stiffness was calculated using the dynamic elastic modulus formula of the small-strain soil. The test results showed that the scour depth identified using the CMCM method was conservative, with an accuracy of 80%.

Some studies constructed explicit or implicit relationships between scour indicators and scour depth using appropriate samples, and then invert scour depth based on measured structural vibration data. They can be viewed as a simplified surrogate model approach.

Xiong et al. [56] proposed an inversion method for scour depth of continuous girder bridges. Structural flexibility matrix D was derived using the frequency and mode shape of the bridge superstructure, and the D-based deflection change Δδ was defined as a scour indicator. The relationship between Δδ and scour condition of the tested bridge was established via a numerical simulation, and the scour depth was obtained by substituting the measured Δδ on site into the relationship formula. A numerical case of a three-span continuous girder bridge was used to verify the methods. Only the first transverse bending mode of the superstructure sensitive to scour was used to calculate the Δδ, and the results show that Δδ increased with the increase in scour depth. Twenty-one scour samples within 5 m scour depth were obtained using the FE model, and the polynomial function relationship between Δδ and scour depth was further fitted. The numerical results showed that the fitting relation can invert the scour depth of single and multiple piers with high precision.

Bao et al. [93,94] proposed using the bridge scour characteristic curve (BSCC) to predict the scour depth of pile foundation. BSCC represents the relationship between pier frequency and scour depth. Through theoretical, numerical and laboratory test studies, it is proven that for piers partially embedded in the soil, using the Winkler model to simulate soil–structure interaction can more accurately predict pier frequency than the Pasternak model [93]. Further, a BSCC prediction method based on the Winkler model and global optimization technology was proposed and was verified via numerical simulation [94]. The results showed that accurate BSCC can be obtained using only a small number of samples (2~4) within a small scour depth (0.2~0.5 m). The BSCC can accurately predict the scour depth of pile foundation and the subsoil reaction modulus of the bottom soil layer that has the greatest influence on pier frequency, regardless of single or multiple soil layers. The proposed method does not require a FE model and provides a feasible method for frequency-based scour depth prediction.

In conclusion, the effectiveness of using model updating to identify bridge scour depth has been verified via numerical and experimental studies. For long-span bridges, the dynamic characteristics of superstructure were used to update the model. They are easily accessible from BSHM systems, but the influence of environmental conditions on them need to be considered. For middle-small span bridges, pier frequency was generally used to update its model. Due to the large number of such bridges, future work will require efforts to improve efficiency and reduce costs. Both scour and soil properties will affect structural dynamic properties. Accurate prediction of soil–structure interaction (SSI) was the premise of scour depth prediction, but it was still a challenging work.

4.2. Scour Depth Identification Based on Machine Learning

In recent years, civil engineering technology tends to be rather efficient with the development of computer and sensor technology [95]. In the fields of structural damage detection (SDD), machine learning (ML) and even deep learning (DL) have been widely used [96]. Avci et al. [97] comprehensively reviewed the development process of civil structure SDD from traditional methods to ML and DL methods, and pointed out that ML method was superior to traditional methods in processing fuzzy and noise-disturbed data. Recently, there have been some explorations of using the ML method to identify scour. The basic idea was to construct a ML model for scour depth identification with structural dynamic characteristics as input and structural state as output. Those methods were expected to provide higher accuracy and efficiency under the influence of complex factors. The main progress is reported below.

Feng et al. [98] used support vector machines (SVM) to identify bridge scour depth and evaluate bridge safety. The proposed method was verified via numerical simulation. A finite element model of a simply supported beam bridge with a simple soil-spring model was constructed to calculate the bridge frequency, and was verified via field tests. The influence of scour and design parameters of bridge substructure on bridge transverse frequency was analyzed comprehensively. Thus, seven important factors were determined, such as soil property, pile length, pier height and pier diameter. According to the ratio of scour depth h to pile length L, four safety levels of bridges were defined, namely safety (h < 1/4L), warning (1/4L < h < 1/3L), danger (1/3L < h < 1/2L) and serious danger (h > 1/2L). A support vector machine (SVM) model for bridge safety assessment was trained using the cross validation method, with seven factors and bridge frequency as input and bridge safety level as output. A case with 770 training samples and 126 test samples was used to verify the SVM model, and the results showed that the SVM model had a conservative prediction of 75% and thus can preliminarily judge the bridge safety.

At present, the main process of identifying bridge scour depth based on ML method is as follows. Firstly, representative scour case samples for the tested bridge were provided using the finite element method. Secondly, the ML model was trained with critical bridge frequency as input and scour depth as output. Finally, the bridge frequency identified via the field vibration test was put into the training model to predict the scour depth. This process has been widely applied in cable-stayed bridges. For example, using the frequency of bridge superstructure, Zhang et al. [99] trained a SVM model and Xiong et al. [100] trained a BP neural network model. Numerical results showed that those models have high prediction accuracy.

The influence of environmental temperature on scour identification can be mitigated by using the ML method. Zheng et al. [101] constructed a Gaussian Process (GP) model to describe the correlation between modal properties of a monitored bridge and temperatures, and used the model to mitigate the influence of temperature on the modal properties. On this basis, the bridge scour depth was inferred using the Bayesian inference methods. The effectiveness of the proposed method was verified via numerical simulation. The sample data required to train the GP model was provided via the bridge finite element model, and temperature changes were indirectly simulated by changing material properties and constraint stiffness of expansion joints. The numerical results showed that the GP model can effectively eliminate the influence of temperature, and the modified modal properties were basically consistent with the analysis results of the FE model. The scour depth inferred by modified modal properties was more accurate than that by original modal properties. The maximum error of the former was only 11.28%, while that of the latter was more than 20%.

Foundation scour also threatens the safety of offshore wind turbines. The ML method was also used to identify the scour depth of such structures. For example, Wang and Jiang [102] proposed a method to identify scour depth and damage of wind turbines based on One Dimensional Convolutional Neural Network (1D CNN). The method was verified via numerical simulation. The CNN was trained with the frequency of the structure as input and the structural damage stats (including three types of scour, structural damage, scour and structural damage) as output. The results showed that the CNN can accurately estimate the scour depth and identify the damage location and degree, even if the scour and structural damage existed simultaneously.

In conclusion, the ML method can identify scour depth more directly and intelligently without relying on complex structural analysis theory. This method has great advantages in solving complex coupling problems, such as environmental factors interference and multi-damage coupling. However, the training of ML models required a sufficient number of representative samples, which were usually obtained using the finite element methods, resulting in a long time to prepare the samples and train the models. Therefore, ML-based scour identification method may be more valuable in large-scale middle-small span bridge groups.

5. Discussion

Based on the above review, some important issues in vibration-based scour diagnosis are summarized and discussed in depth in this section, which are related to foundation types, bearing conditions, soil–structure interactions and excitation modes, respectively. Further, some possible research trends are proposed.

5.1. Influence of Bridge Foundation Type on Scour Diagnosis

The influence of scour on the bridge dynamic performance varies with the type of foundation, leading to different scour indicators. For shallow foundations, scour removes the soil around the foundation, weakening the horizontal constraint of the soil for the foundation and the stress level of the basement soil, resulting in a decrease in the horizontal and vertical stiffness of the substructure [42,72,73]. Therefore, scour indicators can be constructed using bridge horizontal and vertical modes or vibration. For the deep foundation, the foundation is constrained horizontally and vertically by the surrounding soil in a large depth. Scour weakens the lateral constraint of the upper soil on the foundation and results in the decrease in the bridge horizontal frequency [69,70,71,78]. Compared with shallow foundation, the vertical dynamic characteristics of bridges with deep foundation have little change unless severe scour occurs [39], so scour indicators can be constructed using bridge horizontal modes or vibration.

It is noteworthy that scour directly weakens the stiffness of the substructure and indirectly affects the dynamic characteristics of the superstructure. The sensitivity of local pier frequency to scour is significantly higher than that of the overall bridge frequency [71,72,78], but it is more convenient to test the superstructure frequency than the pier frequency, which has led to differences in the selection of scour indicators in existing studies. In conclusion, a two-stage scour diagnosis scheme is proposed. Firstly, a preliminary rapid scour diagnosis is performed through vibration tests on the superstructure to infer the possible scour location, during which sensors are installed on the bridge deck and even passing vehicles. Then, detailed scour diagnosis is performed through vibration tests for abnormal piers, during which sensors are installed on the piers using temporary work platforms, robots, etc.

5.2. Influence of Bearing Damage on Scour Diagnosis

The influence of bearing damage on scour diagnosis has not been paid much attention. During bridge operation, bearing diseases, such as emptying and aging, occur frequently and affect its working performance. The three-dimensional stiffness of the bearing affects the dynamic characteristics of piers and bridge deck. Scour and bearing damage may occur simultaneously, and the author’s preliminary study showed that the influence of the two damages on the bridge dynamic behavior can reach the same level with a significant coupling effect. Xiong et al. [45] proposed two measures to reduce the influence of bearing damage on bridge scour identification. First, structural anomalies during floods were considered to be caused by scour, regardless of influence of bearing damage. Second, the influence of the bearing was mitigated by detecting the bearing condition. However, these two measures cannot completely solve the problem, because bearing damage is sometimes concealed and difficult to detect visually. The bearing condition may change under the time-varying impact of traffic load and environmental factors. Importantly, only after the bearing condition is determined, it is possible to identify scour depth using model updating [58]. In the future, it will be necessary to extensively study the coupling effects of bearing damage and scour on the structural dynamic behavior, and propose a damage diagnosis method for bridge substructure considering multi-damage coupling.

5.3. Influence of Soil–Structure Interaction on Scour Diagnosis

Accurate simulation and prediction for soil–structure interaction (SSI) is a crucial prerequisite for identifying scour depth. In existing studies, the linear Winkler beam model has been extensively used to simulate SSI, and its effectiveness has been verified through field and laboratory tests [71,77,80,93,98]. Although a variety of methods, such as the small strain stiffness method (SSSM) and the American Petroleum Institute method (APIM), have highlighted the stress–strain relationship of soil, it is still a great challenge to accurately calculate the soil spring stiffness without prior information about soil deposition conditions. Some studies used a few reference samples with known foundation state and corresponding structural frequency to correct the soil spring stiffness calculated using the design parameters [41,58]. However, because of the uncertainty and variability of soil properties along the depth direction, this kind of inverse problem may have multiple solutions due to insufficient constraints. Moreover, it is difficult to obtain reliable reference samples in the field and it is not possible to verify the correction results of soil spring stiffness. Since both soil properties and scour affect dynamic characteristics of bridges, a reasonable strategy is to first determine soil properties and then quantify scour depth. This strategy has been applied in scour identification based on model updating [41,43,58,92], but the process was complex and time-consuming, and it was more valuable for long-span bridges. Therefore, an important trend for the future is to propose a rapid SSI prediction method for middle-small span bridges, and one promising strategy is to develop intelligent prediction models using ML methods.

5.4. Innovative Excitation Methods on Scour Diagnosis

The existing research focused on vibration signal processing but ignored the evaluation and improvement for excitation methods. The applicability of different excitation methods is discussed based on the above review, and some ideas about innovative excitation are presented.

Most of the field piers with scour risk are located in complex deep-water environments. The application of forced excitation requires the installation of temporary working platforms and vibration exciters by boats, bridge inspection vehicles, etc., which is costly and time-consuming [41]. This indicates that forced excitation is not suitable for field scour detection.

Hydraulic excitation is the natural lateral force that can excite pier vibration in the flow direction. Currently, the potential of hydraulic excitation in scour diagnosis has been verified via numerical simulation and laboratory tests [47,52,75,78], but there are only a few field experimental studies. Many problems need to be solved urgently. First, the vibration modes of piers excited by hydraulic excitation varied with the type of foundation and pier [55], and no uniform law was obtained in field tests. Second, as a time-varying excitation, unsteady input of hydraulic excitation and the induced weak vibration output make vibration measurement difficult. Third, environmental conditions, such as temperature and random noise, have obvious interference to scour diagnosis.

Moving vehicle excitation is commonly used in the scour diagnosis for middle-small span bridges. The vertical vibration of bridge deck is easily stimulated by wheel force but the horizontal vibration is difficult to be excited [78]. Especially for the bridge with the deep foundation, the vibration signal of piers induced by wheel force has small amplitude and low signal-to-noise ratio, which are easy to be interfered by the field test conditions, such as speed and road roughness. Innovative excitation methods may break the above difficulties. For example, based on the fast and flexible braking function of the vehicles, the braking force was used to excite the bridge longitudinal vibration, and its effect is better than that of the normal traffic load [72]. Several potential methods for rapid scour diagnosis have been proposed and validated via numerical simulation [73,74], and laboratory and field tests are also being carried out. In addition, moving impact vibration test technique, with the advantages of both forced excitation and moving vehicle excitation, may also be used to develop innovative, reliable and rapid scour diagnosis schemes.

6. Conclusions

Vibration-based scour diagnosis methods have great potential for rapid, low-cost scour diagnosis and structural integrity assessment, overcoming the difficulties in the field installation and operation of traditional equipment. A systematic and hierarchical review was provided in this paper. Firstly, the principles of scour diagnosis and vibration excitation method were introduced. Then, existing qualitative and quantitative studies were reviewed. The former referred to identifying scour locations based on changes in structural dynamic characteristics and dynamic response, and the latter referred to quantifying scour depth using model updating or machine learning methods. On this basis, some important but neglected issues were summarized and discussed. The conclusions and future prospects are summarized as follows:

• The qualitative scour diagnosis includes two interrelated levels. First, bridge vibration is excited by a fast and effective excitation method. Second, scour indicators with high noise resistance and high damage sensitivity are extracted from the recorded vibration data. Future work needs to concentrate on the third level, namely the evaluation of structural safety conditions based on scour indicators, to provide the basis for the next decision.
• The scour depth can be identified using model updating and machine learning techniques. The complicated theory and process of model updating result in its low efficiency, and it cannot identify scour conditions in real time. Machine learning technique is a more promising approach, which may realize long-term real-time scour monitoring by constructing an effective scour prediction model or digital twin model for long-span or middle-small span bridges.
• Some critical issues are not being addressed. Bearing conditions, soil properties and scour condition jointly determine the constraint conditions of bridge substructure and affect the structural dynamic behavior. A critical prerequisite for identifying scour depth is to determine two other factors. However, there are only a few research on bearing damage, and the SSI prediction is still a challenge. Moreover, the time-varying temperature has a significant effect on scour diagnosis as it changes the mechanical properties of bearings, expansion joints and concrete.
• For the large number of middle-small span bridges, a fast and low-cost scour detection method is urgently needed. Some feasible schemes were proposed by using traffic-induced vibration and improved brake-induced vibration. Indirect detection based on vehicle response is also worth exploring. Using machine learning and data fusion technology to analyze the big data of vehicle response is expected to quickly detect the scour condition of bridge groups.