One of the most prevalent causes of bridge failure around the world is “scour”—the gradual erosion of soil around a bridge foundation due to fast-flowing water. A reliable technique for monitoring scour would help bridge engineers take timely countermeasures to safeguard against failure. Although vibration-based techniques for monitoring structural damage have had limited success, primarily due to insufficient sensitivity, these have tended to focus on the detection of local damage. High natural frequency sensitivity has recently been reported for scour damage. Previous experiments to investigate this have been limited as a result of the cost of full-scale testing and the fact that scaled-down soil-structure models tested outside a centrifuge do not adequately simulate full-scale behaviour. This paper describes the development of what is believed to be the first-ever centrifuge-testing programme to establish the sensitivity of bridge natural frequency to scour. A 1/60 scale model of a two-span integral bridge with 15 m spans was tested at varying levels of scour. For the fundamental mode of vibration, these tests found up to a 40% variation in natural frequency for 30% loss of embedment. Models of three other types of foundation, which represent a shallow pad foundation, a deep pile bent and a deep monopile, were also tested in the centrifuge at different scour levels. The shallow foundation model showed lower frequency sensitivity to scour than the deep foundation models. Another important finding is that the frequency sensitivity to “global scour” is slightly higher than the sensitivity to “local scour”, for all foundation types. The level of frequency sensitivity (3.1–44% per scour depth equivalent to 30% of embedment of scour) detected in this experiment demonstrates the potential for using natural frequency as an indicator of both local and global scour of bridges, particularly those with deep foundations.

Bridge scour refers to the removal of soil from around structural foundations located in a river or coastal region as a result of the erosive action of water [

In contrast to other available techniques [

Vibration-based scour monitoring has shown considerable potential when analysed with computational models of bridge-soil systems [

Establishing the sensitivity of natural frequency to scour in practice is difficult—a monitoring system would need to be installed on a candidate bridge, with no guarantee of measuring any scour within the timespan of the project. On the other hand, full-scale testing of controlled scouring of a bridge is not viable because of the costs involved, and tests on scaled-down soil-structure models at normal gravity [

Another important consideration is whether or not natural frequency is less sensitive to “local scour” at a bridge foundation than to “global scour”. As shown in Fig.

Shape of a local and global scour

This paper describes the development of a centrifuge experimental programme to capture natural frequency variation of bridges due to local and global scour. The objectives of the research were to:

Develop a scale-model testing methodology for assessing vibration-based scour monitoring techniques;

Identify the natural frequency sensitivity to scour of different bridge types;

Establish the natural frequency sensitivity of bridges due to different forms of scour, i.e. local and global scour.

Centrifuge modelling is a technique used to correct inaccuracies in soil property scaling that is involved when attempting to model a full-scale (also called prototype-scale) soil mass with a small-scale (also called model-scale) soil model. To understand this scaling inaccuracy in the context of vibration behaviour applicable to this research, consider a 1/

Properties of full-scale and 1/

Any full-scale depth

Equations _{s}) at _{f}) at the corresponding full-scale depth (

As soils exhibit highly non-linear stress-dependent behaviour, the correct behaviour will only be observed if the stress levels in the small and full scales match. Equal stress condition, that is,

Stress similitude between small and full scale at a length-scale factor of

According to Eq. (

All tests for this research were conducted using the Turner Beam centrifuge at the Scofield Centre, University of Cambridge. This is a 10 m diameter centrifuge capable of subjecting models with a mass of up to 1000 kg to centripetal accelerations of 125

This centrifuge testing programme modelled the soil-structure interaction with a package of 434–485 kg in mass. It was tested under a 40

These centrifuge tests were intended to model the change in dynamic response of various types of full-scale bridges using small-scale models. The constraints in the centrifuge model container and typical field conditions required iterative selection of the full-scale and small-scale properties. Only the finally selected properties and applicable constraints are given here.

Three hypothetical full-scale bridges were selected to compare the viability of the vibration-based monitoring method in different types of bridges. The key differences and similarities among these bridges are listed in Table

The three types of full-scale bridges considered

The full-scale Bridge 1 was considered to be an “integral” bridge. Integral bridges have low maintenance costs because they have a monolithic connection at superstructure-substructure connections, without any bearings or expansion joints. Highways England recommends integral bridges as the first option to be considered for bridges with lengths not exceeding 60 m, skews not exceeding 30°, and settlements that are not excessive [

The full-scale Bridges 2 and 3 considered were both “simply supported”. A significant portion of the existing concrete bridge stock around the world also have simply supported bridge decks [

The full-scale Bridges 1 and 2 were both selected to have pile bent foundations (a row of piles from deck level to the bottom of the foundation). This pile bent foundation arrangement included four piles of 18 m in length, 12 m of which was driven into the soil, with the remaining 6 m above ground level. The pier piles were assumed to be 760 mm in diameter. The abutment piles in Bridge 1, the integral bridge, were selected to be 540 mm in diameter. The abutments of Bridges 2 and 3, the simply supported bridges, were assumed to behave as fully rigid supports.

Bridge 3 was selected to have a shallow pad foundation. The shallow foundations are economically viable in bridges typically when the required soil properties are present within 3–4.5 m from ground level [

The maximum span of beam-slab-type bridges is between 10 and 30 m in the majority of bridges in Europe [^{–2} m^{4} and a cross-sectional area of 0.31 m^{2}. The spacing between Y1 beams was chosen as 1.5 m and the in situ slab depth was chosen as 0.2 m, based on the typical details given in a precast beam catalogue by Concast Precast Ltd (2009) [

These full-scale hypothetical bridges were considered to be newly concreted bridges with a history of low stress levels, without significant cracking. Thus, all full-scale bridge elements constructed of C40 grade were assumed to have a modulus of elasticity of 35 GPa and a bulk density of 2550 kgm^{−3} [

Figure ^{5} kg in full scale, which was assumed to represent the extra mass of the three diaphragm beams in addition to the mass of the overlapping deck. The composite beam-slab deck in the full-scale Bridge 1 was simplified to a rectangular-slab bridge deck, having the representative flexural rigidity and mass in small-scale Model 1.

Scaling down of three full-scale bridge types to four small-scale centrifuge models

The presence of bearings in full-scale Bridges 2 and 3 complicates the testing of simply supported bridges in small-scale experiments. Therefore, only the “standalone foundations” (foundation-only models) of the bridge piers of Bridges 2 and 3 were scaled down as centrifuge models, with the aim of representing the local bending modes of the foundations that do not involve any deck vibration and thus can be represented by standalone foundations.

As shown in Fig. ^{5} kg at the top.

As shown in Fig.

All lengths of the elements were scaled down by _{0}) should be correctly scaled down by

All model structures were made from aluminium alloy elements. As a result of the change of material properties from full scale (concrete) to model (aluminium), not all of the centrifuge scaling laws could be met at the same time. Only the most critical parameters required were therefore selected to model the mass and stiffness properties needed to simulate the dynamic behaviour of the full-scale bridge. The correctly scaled critical properties are shaded in Table ^{−1}, but the pile alone provides only 0.09 kgm^{−1} (71% error), whereas the pile filled with soil provides 0.23 kgm^{−1} (25% error). For the elements above ground level, such as the deck and columns, the flexural rigidity and mass per unit length were chosen as the main parameters for scaling, as there is no soil-structure interaction.

Model element selection (shaded properties adhere to the scaling laws)

Based on the small-scale properties derived in the previous section, Models 1–4 were constructed in a soil medium in a cylindrical centrifuge container. Figures

Experimental set-ups, sensor arrangement and scour levels corresponding to all the models in the centrifuge container—all dimensions are in mm

The cylindrical centrifuge container with all four structural models and the excitation set-up

The model structures were made from aluminium sections and had the small-scale properties listed in Table

The circular hollow piles in the model structures were made of aluminium alloy 6061-T6, and the rest of the solid sections were made of aluminium alloy 6082-T6. These aluminium alloys have Young’s modulus of 70 GPa and density of 2700 kgm^{−3} [

All of the model structures had at least one integral connection, either at the foundation to deck connection or at the stub column to pad foundation connection. To obtain these integral connections, sockets of the same cross-sectional area as the column/pile elements were first machined up to half the depth of the slab/pad section. The columns/piles were then inserted into these sockets and glued in place with a high-strength retainer adhesive. The connection rigidity of each pile/column was tested by finding the fundamental sway natural frequency of the piles when the slab/base was fixed. The piles without adequate integral connection showed a varying frequency for repeated tests or a lower frequency than other similar piles. These piles were removed and reconnected to obtain the desired integral connection.

The accelerometer locations on Model 1 are shown in Fig.

The soil model was prepared in a cylindrical steel container, as shown in Fig.

The soil used for the centrifuge test was Hostun sand acquired from Drôme in the south-east of France. This sand type is widely used in centrifuge model tests. It is a high silica (SiO_{2} > 98%) sand of grain shape varying from angular to sub-angular [

Geotechnical properties of Hostun sand [

The sand was poured with an automatic sand pourer developed by Madabhushi et al. at the Scofield Centre centrifuge testing facility [

After a calibration sand-pouring test, a drop height of 690 mm through a 5 mm diameter nozzle (no sieve), with a 20 mm spacing between consecutive motions, was selected to obtain a relative density of 66%. The relative density of the final sand model in the centrifuge container was found using the pre-post weight difference and the properties given in Table ^{−3}.

The sand pouring was paused when the sand fill was at 175 mm to place Model 3 with the shallow foundation, and then the pouring continued to embed the foundation.

The structural models with pile foundations (Models 1, 2 and 4) were inserted into the soil model at the end of the sand pouring. The piles were pushed down until the desired embedment depth of 200 mm had been reached. It was assumed that the piles with open ends would not cause significant disturbance to the soil when inserted at normal gravity.

Two excitation sources were used to test two models during each centrifuge flight. A piezoelectric actuator, which had previously been used to test monopile foundations [

The amplified piezoelectric actuator used was an APA400MML [

Amplified piezoelectric actuator set-up

An automatic modal hammer was developed to excite Models 2 and 3, as it was thought to be able to provide a better external excitation than the piezoelectric actuator, which rests on the model. As shown in Fig.

Set-up of the automatic modal hammer controlled by air-pressure

All data acquisition was carried out with the centrifuge data acquisition system, which uses slip rings to transfer data to the computers in the centrifuge control room. All signals were transmitted through a set of amplifiers and then through an analogue to digital converter [

Before each centrifuge flight started, the desired scour hole was created using vacuum suction to limit disturbance to the underlying soil. The depths of the scour hole were measured with reference to the paper rulers that were on all of the structural model foundations. Figure ^{0} was maintained around all sides (Fig.

A local scour hole created at the pier foundation of Model 1

Following the creation of the scour holes, the natural frequencies of the models were measured at normal gravity, 1

The different steps of scour tested around each centrifuge model are summarised in Table

The scour steps around the bridge models

With Model 3, local scouring was introduced in 17 mm steps (0.68 m in the full scale), with a final 50 mm (2 m) global scour case, requiring a total of five centrifuge flights. With Models 2 and 4, local scouring was introduced in two 30 mm steps (1.8 m) up to 60 mm (3.6 m), and a final global scour of the same depth was tested, making a total of four centrifuge flights.

The air temperature during the one-week centrifuge flight programme was measured at the end of every centrifuge flight. The temperature was stable (12.5–16 °C) throughout the experimental programme, as the centrifuge is located underground. Therefore, it was assumed that there were negligible temperature effects on the natural frequencies of the models, and they varied entirely as a result of scour.

For each scour level, the soil around the piles provides some stiffness to the models. The maximum soil stiffness, and thus the maximum natural frequency, is reached when the embedded layer of soil provides full fixity. Therefore, as shown in Fig.

Fixed-base test being carried out for Model 1

The logged input and output data were analysed in MATLAB. An example input excitation, and the corresponding response measured in Model 1, are shown in Fig.

Measured input excitation and the response acceleration of one sensor in Model 1 (scour step M1S7b in Table

Modal analysis for Models 1 and 4 was therefore carried out using the output-only method, frequency domain decomposition (FDD). The FDD method is based on the singular value decomposition of the measured PSD matrix

Figure

First singular value of Model 1 response for the same scour case but with two excitation sources: no actuator (step M1S7a), and with actuator (step M1S7b)

A typical sway mode shape of Model 1—side view (“X”—measurement locations)

Models 2 and 3 were excited by impacts from the automatic modal hammer, and the natural frequencies found from a simple frequency–response function (FRF). The FRF quality was improved by segmenting the signals by individual impacts and applying a force window to each segment of input and an exponential window to each segment of output [

Time histories of Model 3—scour step M3S5

Accelerance FRF and coherence for Model 3 (scour step M3S5 in Table

As the environmental parameters were controlled, any change in natural frequency can be attributed to the scour itself. Therefore, high sensitivity of natural frequency indicates a high potential for it to be used as an indicator of scour, while low sensitivity indicates otherwise.

The spectra obtained from modal analysis showed reliable modal peaks only for the fundamental modes, and thus all of the natural frequencies discussed here correspond to the fundamental sway mode (Fig.

For comparison, the natural frequencies measured at “1

The full-scale natural frequency variation found using the integral bridge model is shown in Fig.

Representative full-scale natural frequency variation of integral bridge

As shown in Fig.

An additional 3.6 m of local scouring at the left abutment caused a frequency reduction of 0.21 Hz (− 14%) from 1.53 Hz. Furthermore, 3.6 m of local scouring at the right abutment resulted in a 0.18 Hz (− 14%) reduction in natural frequency from 1.32 Hz. Both of these local scour cases at the abutments represent approximately 4% of natural frequency shift for every metre of full-scale local scour at one of the abutments. The cumulative effect of 3.6 m deep local scour at each of the three supports resulted in an overall 38% reduction in natural frequency.

As shown in Fig.

The full-scale natural frequency variation of the pile bent foundation is shown in Fig.

Full-scale natural frequency variation of pile bent

The initial 1.8 m of local scour caused the natural frequency at 60

The total 3.6 m of local scour gave a 41.4% frequency reduction. This reduction corresponds to an average natural frequency sensitivity of 11.5% per metre of full-scale local scour.

The global scour of 3.6 m caused the natural frequency of the model to fall to 1.66 Hz, lower than the representative local scour value of 1.73 Hz. However, the overall natural frequency sensitivities due to the local and global scour were not significantly different. The overall natural frequency sensitivity to global scour of 3.6 m was 43.7%, and due to local scour of 3.6 m was 41.4%—only a 0.6% sensitivity difference due to 1 m of scour.

The variation of the natural frequency with scour depth of the shallow foundation is shown in Fig.

Full-scale natural frequency variation of shallow foundation

A local scour hole of 2 m at full scale was created in three equal steps around the column of the shallow foundation. The full 2 m of local scour caused the original frequency of the model (2.44 Hz) to fall to 2.27 Hz—on average, a 3.5% frequency shift for every metre of full-scale scour depth. Similar to Model 2, the natural frequency sensitivity of Model 3 showed reduced sensitivity with deeper local scour depths. For example, the frequency changed by 4.7% and 2.3%, respectively, for the first and second metre of local scour steps.

The natural frequency of Model 3, measured with equivalent full-scale global scour depth of 2 m (2.21 Hz), was slightly lower than the natural frequency measured with 2 m of local scour (2.27 Hz). The global scour generated an overall frequency shift of 9.4%, which equates to 4.7% per metre of equivalent full-scale scour depth. Therefore, global scour shows 1.2% higher frequency sensitivity than local scour for every 1 m of full-scale scour. This higher frequency sensitivity is to be expected, as unscoured soil, slightly distant from the foundation, still provides a degree of increased pad foundation fixity, leading to higher stiffness and hence natural frequency for local scour relative to global scour.

Figure

Full-scale natural frequency variation of monopile foundation

The full-scale local scour of 3.6 m around the monopile caused the fundamental natural frequency to fall from 4.15 to 3.24 Hz. This represents a 22% change in natural frequency, equivalent to approximately 6.1% per metre of full-scale scour.

Global scour of 3.6 m at full scale represents a natural frequency change of 22.9%, equivalent to approximately 6.4% per metre of scour. Global scour of 3.6 m caused the natural frequency to reduce slightly more (0.04 Hz) than the local scour of 3.6 m. However, the difference between the global- and local-scour-induced frequency shifts was only 0.3% per metre of scour.

All foundation types exhibited a measurable change in natural frequency due to scour. A summary of the measured sensitivities is shown in Fig.

Summary of frequency sensitivities at full scale

Current numerical models of soil-bridge systems that aim to simulate scour are limited by the representation of the soil-foundation interface. These use Winkler spring models to represent the soil small-strain stiffness, and scour is simulated by simply removing springs down to the level of scour, without making any distinction between local and global scour [

Figure

Some limitations of this research are worth noting. The fundamental vibration mode captured with Models 2, 3 and 4 (which have free ends with no restraints at the top) does not directly represent the fundamental mode of a simply supported bridge with the deck and bridge bearings of Bridges 2 and 3. With these additional stiffnesses and masses, the frequency sensitivity of a simply supported bridge may differ from the results reported here. However, the finding that deep foundations are more sensitive to scour than shallow foundations would still hold true in Bridges 2 and 3, since the introduction of the same deck arrangement can be expected to have a similar effect to the two foundation types. This aspect needs further investigation using numerical models of these foundation models, which can be calibrated using the experimental results observed here for local and global scour as explained in Kariyawasam [

Another limitation is that the soil model in the centrifuge experiment was of uniform density, whereas, in practice, density may vary with depth in different layers of soil. It is not clear to what extent this may be significant, especially in the case of piles where the pile-head stiffness is dominated by the soil restraint around the pile head (near the ground level), rather than the restraint offered at depth. A gradual reduction in frequency sensitivity with scour depth may not be noticed with a real bridge on layered soil. This aspect needs further investigation, with enhanced models to represent layered soils.

The centrifuge test was conducted in dry soil with no water present but the saturated or partially saturated soil present in a typical field-scale bridge may slightly change the frequency sensitivities to scour. However, such a change is unlikely to be significant, as previous numerical and experimental research suggests that the effect of water is negligible on the natural frequency sensitivity to scour of typical stiff bridge piers [

This paper has described the development of a centrifuge testing methodology to measure, for the first time, natural frequency sensitivity to local and global scour at the foundations of different bridge types. A scaled-down two-span integral bridge model and standalone foundations of two simply supported bridges were constructed in the sand with 66% relative density. These structural models were excited with a piezoelectric actuator and an automatic modal hammer developed for this test. The automatic modal hammer was found to have provided better impulsive excitations than the piezoelectric actuator.

The fundamental natural frequency of the integral bridge reduced by up to 40% due to scour depths equal to approximately one third loss of pile embedment. These sensitivities are significantly higher than the reported natural frequency sensitivities to considerable structural damage and environmental sensitivities (0.4–7%) [

The fundamental natural frequencies of standalone foundation models produced the highest sensitivity to a one-third loss of embedment for the pile bent (44%), the second highest for the monopile (23%) and the lowest for the shallow foundation (4%). While these fundamental modes are not directly representative of the simply supported bridges with deck and bearings interactions, they do show that deep piled foundations have significantly greater natural frequency sensitivities than shallow foundations. Once the bridge deck is included to these standalone foundation models, the natural frequency sensitivities to scour could be lower than reported here since then the loss of soil-structure interaction because of scour becomes only a fraction of the overall stiffness provided by bridge deck, bearings and soil. Therefore, natural frequency is unlikely to be a reliable indicator for a simply supported bridge with shallow foundations. These experimental results for standalone foundations could be extrapolated to the simply supported bridges using numerical modelling in future studies.

The natural frequency reduction due to global scour was greater than for local scour; however, the difference was only 1–2% for one-third loss of embedment. Numerical modelling techniques could be developed in future studies to simulate these different effects of local and global scour. The frequency sensitivity to local scour tended to reduce as the scour hole deepened, so natural frequency as a measure of the progression of local scour may be more suitable for a bridge with no existing scour than for a bridge with severe local scour already present.

This study has demonstrated that natural frequency has significant potential as an indicator of scour for bridges with deep foundation/integral bridge decks. It is recommended that future studies consider further the influence of soil layering, as well as the likely natural frequency shifts due to other environmental factors such as temperature and water level, such that a detailed methodology for scour monitoring may be developed to implement this technique in practice.

We are grateful to the Cambridge Centre for Smart Infrastructure and Construction (Innovate UK Grant reference number 920035 and EPSRC Grant no EP/N021614/1), and the Laing O’Rourke Centre for Construction Engineering and Technology, for the funding provided for this experimental programme. The PhD funding provided by the Gates Cambridge Trust, founded by the Bill and Melinda Gates Foundation (Grant number OPP1144), is also gratefully appreciated. The valuable support provided by Tom Williams during the planning stage of this project is acknowledged. The research facilities provided by the Schofield Centre are gratefully acknowledged. We would like to acknowledge the initial development of the air hammer set-up by Peter Kirkwood and the piezoelectric actuator set-up by Jing Dong for testing monopiles. We thank Deryck Chan and Jad Boksmati for the valuable guidance provided to run the centrifuge test. We would also like to thank Geoff Eichhorn, Ahmed Alagha and Thejesh Garala for their guidance during the model preparation. Finally, thanks to Scofield Centre senor technicians Kristian Pether, Chris McGinnie, Mark Smith, John Chandler and David Layfield for their assistance at different stages of the experiment. Data supporting this paper is available from the University of Cambridge Open Data repository [

^{®}Accelerometers ADXL78

Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.