Microwave Structural Health Monitoring of the Grouted Connection of a Monopile-Based Offshore Wind Turbine: Fatigue Testing Using a Scaled Laboratory Demonstrator

,


Introduction
Ofshore wind energy has emerged as a vital source of renewable power in recent years.In 2022, the nominal global ofshore wind power capacity increased by 8.8 GW, bringing the total capacity to 64.3 GW [1].Forecasts indicate that the market will grow signifcantly between 2023 and 2027, with an expected 130 GW to be added worldwide during this period.In Europe, the cumulative ofshore wind capacity reached already 30 GW. Te European Union (EU) has outlined in its ofshore renewable energy strategy [2] that it aims to reach an installed capacity of at least 60 GW by 2030 and 300 GW by 2050.
As ofshore wind farms grow in size, the need for robust monitoring systems increases as they help ensure the safety, reliability, and performance of ofshore wind turbines, particularly with regard to foundation structures.Te benefts of a structural health monitoring (SHM) system for the support structures of ofshore wind turbines are summarized in [3], considering relevant failure modes.
Te prevailing choices for ofshore wind turbine foundations are monopiles [4].After the monopile has been driven into the seabed, the transition piece with a larger diameter is placed on top with an overlap of several meters.Te gap between both steel tubes is flled with highperformance grout, a special type of concrete, forming the grouted connection (see Figure 1).During the operation, the ofshore wind turbine is subjected to harsh weather conditions and intense dynamic loads caused by wind and waves.It has been observed in [6] that the grouted joints are susceptible to cracking due to the large bending moments they experience.Te presence of cracks enables the penetration of water, leading to the possibility of additional crack propagation [7] and erosion [8].
Tziavos [9] describes that, in September 2009, regular inspections of monopiles identifed the slippage of numerous transition pieces by as much as 90 mm.Te following examinations concluded that approximately 700 out of 1,000 monopiles across Europe had slipped signifcantly.A gap between grout and steel was detected at the top of some grouted joints and water ingress was evident in certain cases.Te design practices for grouted connection on ofshore wind turbines were consequently reevaluated.Tis highlights the importance of detecting any changes in the grouted connection as early as possible to ensure the stability of the ofshore wind turbines.Terefore, methods are necessary that can provide continuous and nondestructive monitoring of the grouted connection during operation, ultimately establishing a SHM system.
In the feld of SHM, the importance of speed, accuracy, and labor efciency for the evaluation of vital structures was underscored by Shan et al. [10].In recent years, a typical sensor confguration to monitor ofshore wind foundations consists of acceleration, inclination, and strain sensors, to study vibration amplitudes, natural frequencies, and inclinations [11].Tis approach is shown, for example, by Stuyts et al. [12] who consider accelerometers and fber optical sensors for structural assessment of a wind turbine monopile.Tey also propose an earth pressure sensor as well as a piezo-resistive pressure sensor.Shan et al. [13] showed that damping ratios can be more sensitive to measurement noise and sensor allocation than modal frequencies.Liu et al. [14] analyzed acceleration and inclination sensors for the impact assessment of typhoon "In-fa" on the dynamic properties of monopile ofshore wind turbines.On top, SCADA (supervisory control and data acquisition) data have been recorded along with the sensor data in order to correlate the measurements with the parameters of the wind turbine such as wind speed, wind direction, turbine yaw angle, and turbine power output [12].
On laboratory scale, Müller et al. [15] successfully demonstrated strain sensing via fber Bragg gratings (FBGs) in a large-scale ofshore wind turbine grouted connection specimen.In that work, damage sensitive features have been derived from the FBG signal responses using the Wigner-Ville distribution and energy spectral density.A large-scale laboratory experiment has also been studied by Häckell et al. [16] using inclination sensors and accelerometers.Other approaches for monitoring grouted connections are based on guided ultrasound waves [17] and electromechanical impedance spectroscopy [18].
Ground penetrating radar (GPR) is commonly used for the nondestructive testing (NDT) and monitoring of concrete structures such as bridge decks, beams, columns, slabs, and walls [19][20][21][22].In many applications, the focus is on the analysis of reinforcements.Manual scanning by moving the radar antennas along the scanning object is frequently used, allowing for the detection and localization of embedded objects, such as rebars and cables.Depending on the radar system parameters (e.g., frequency and transmission power) and the material to be examined, the depth to which faws can be detected varies.Under the conditions described in [23], the detection depth was determined to be up to 20 m at 100 MHz, up to 7 m at 270 MHz and up to 5 m at 400 MHz.Tere are commercial GPR scanners for the examination of steel-reinforced concrete parts [24][25][26].Higher frequencies usually provide a higher resolution and reduced penetration depth while lower frequencies provide greater penetration depth with reduced resolution [27].Te successful application of GPR systems in NDT of concrete structures can be considered as a promising prior work to develop a microwave SHM system for grouted connections.
Te novelty of this research includes the development and testing of a GPR system specifcally designed for the continuous monitoring of grouted connections.A high radar sensitivity is achieved by using a stepped-frequency continuous wave (SFCW) radar design that supports a long measurement time, which enables a high dynamic range as well as low noise levels.Besides the advantages of SFCW radar systems for static scenarios, our approach has several advantages compared to existing nondestructive testing methods for grouted connections.Radar enables the direct, active, and automatic inspection of the grouted connection in ofshore wind turbines.Typically, damage in these connections is inferred indirectly by examining operational parameters and measured loads.Methods such as acoustic emissions are passive measurements and therefore require a high uptime of the measurement system to not miss the relevant emissions.X-rays, ultrasound, and electromechanical impedance spectroscopy are direct measurements, but they have not yet been utilized automatically for grouted connections in ofshore wind turbines, without the need for on-site personnel.Due to the limited accessibility to ofshore wind energy turbines, given their distance from the coastline and variable weather conditions, only an automatic inspection method can operate continuously.For the installation of the radar system, only the inner part of the transition piece needs to be accessed.Tis location ensures good protection of the system from the harsh environment.
Initial results from the accompanying ofshore installation have been presented in [5,28].Te antennas of the GPR system are permanently attached to the grout structure so that relative changes between measurements from baseline conditions and measurements during the subsequent lifetime of the structure can be evaluated.To evaluate the radar systems damage detection performance, a large-scale fatigue experiment has been developed and 2 Structural Control and Health Monitoring implemented where the radar sensor (including two transmitting and three receiving antennas) have been installed.In addition to the radar system results, this study incorporates measurements gathered from the testing system and other synchronized sensors, such as cameras, accelerometers, strain gauges, and acoustic emission sensors.

Design of the Scaled Laboratory Demonstrator.
To assess the SHM system performance under controlled laboratory conditions, a scaled laboratory demonstrator was constructed (Figure 2(a)) for accelerated life cycle testing under realistic operational loads but without environmental infuences.Terefore, the design of the demonstrator focused on the dimensions of a grouted connection at nearly fullscale and the capability to apply realistic load scenarios.Tis enables the application of fatigue tests and structural dynamic tests and provides adequate space for radar antenna placement during the tests, respectively.Since the radar system monitors the grouted connection, it is essential to ensure that any damage, such as cracks, occurs only in the grout during the fatigue tests.Tus, the design and calculation of the structural strength were carried out using the fnite element method (FEM).Tis optimization process aimed to prevent critical stresses from occurring in other parts of the laboratory demonstrator structure (see Figure 2(b)).
Te structure consists of an inner tube (monopile) which measures 813 mm in outer diameter and an outer tube (transition piece) with an inner diameter of 1093 mm.Furthermore, rectangular shear keys with a cross section of 10 mm × 10 mm (see Figure 3(a)) have been welded to both the inner and outer tubes.To facilitate access to the grout with a radial thickness of 140 mm and a height of 1000 mm, the arrangement of the tubes was reversed so that the monopile is facing upwards and the transition piece is fastened to the clamping feld.Tis confguration exposes the upper area of the grouted connection entirely, to give easy access for radar antenna installation and visual inspection of the grouted joint.Additionally, this also provides access to the inner area of the monopile through a ladder.Te grout's bottom is accessible to water and inspection via a 150 mm gap.Te grouted connection utilized a total of 900 kg of PAGEL HF10 high-strength grout [29] to fll the gap between both steel tubes (Figure 3(a)).During several months of curing, the sensor setup, consisting of radar antennas and other sensors, was installed (Figure 3(b)).
Tree hydraulic cylinders are used to apply fatigue loads to the monopile.Tese forces are then transmitted through the grouted connection to the transition piece and the clamping feld.In real wind turbines, the main loading on the grouted connection arises from bending and torsional moments caused by the wind forces at the rotor's height.To simulate the bending moments, the test rig employs two axial 400 kN hydraulic cylinders.Additionally, the shear force is applied by a radial cylinder with a capacity of 100 kN.
Structural dynamic tests were conducted using an electrodynamic shaker (see Figure 4), equipped with an additional excitation mass that was mounted at the top of the laboratory demonstrator to provide the required broadband excitation: sweep sine and random noise in the frequency range from 5 Hz to 1000 Hz.

Description of the Radar
Sensor System.Te radar sensor operates from 100 MHz to 2 GHz in a stepped frequency mode.It allows the simultaneous connection of 2 transmit (Tx) and 3 receive antennas (Rx), alongside a BME280 environmental sensor [30] which is connected to the radar sensor using the serial communication bus Inter-Integrated Circuit (I 2 C).Te air temperature, air humidity, air pressure, and several internal temperatures are recorded.To simplify installation and operation, a single cable provides both power and data connectivity to the sensor via Power-over-Ethernet (PoE).To enable visual detection of fatigue damage and monitoring the laboratory, three cameras are also connected via PoE.Te data collected by all sensors and cameras are frst stored locally before being sent to servers in Frankfurt and Siegen for redundant storage and further analysis.Figure 5 shows the radar sensor for illustration purposes with an open housing cover.Table 1 lists its main characteristics.
Te radar sensor is attached to the outer metal tube using a 3D printed mounting frame that securely hooks onto the tube's edge and is held in place by a screw.At the bottom, the sensor is connected to the tube by a strong magnet.To dampen the vibrations transmitted from the metal tube to Each transmit and receive antenna consists of a pair of fame retardant 4 (FR-4) printed circuit boards afxed to 3D printed PETG components using polyamide screws and nuts.Tese PCBs are soldered to a SMA connector, enabling the antenna's connection to the radar sensor through a coaxial cable of 2 meters length.Te soldering area is protected by liquid rubber before the antenna is flled with PAGEL HF10 grout during the manufacturing process.Tis means that the same grout material is used for the antenna design and the grouted connection of the test rig for the large-scale fatigue experiment.
To install the antennas, the area was marked by self-adhesive foam on the grouted connection, shaping it into the form of a rectangle.Ten, a thin layer of new PAGEL HF10 grout is poured onto the existing grouted joint within this delineated space.Subsequently, the antenna is positioned on this newly laid foundation.

Structural Control and Health Monitoring
Te positioning of each antenna on the grouted connection is depicted in Figure 6 and summarized in Table 2. Based on the fnite element analysis (see Figure 2(b)), antennas Tx2 and Rx2 were positioned equidistantly from one of the areas with the highest stress, at the angular position 180 °.Tis is the region where damage is most likely to occur.Antenna Tx1 was placed at the 0 °position on top of the corresponding area on the opposite side that experiences the same loads.Antenna Rx3 was placed symmetrically between Tx2 and Tx1 so that the distance on the circumference to both is the same.Antenna Rx1 was placed asymmetrically by deliberately moving it closer to Tx1.
Figure 7 presents the measurements of the input refection coefcient S 11 obtained using a vector network analyzer for the antenna characterization.Tese measurements are depicted for two diferent structures: (a) a simplifed structure discussed in [31], utilizing a Keysight M9807A vector network analyzer and (b) the fnal laboratory demonstrator structure, where a NanoVNA V2 Plus4 vector network analyzer was employed two months after the antennas were installed on the grouted connection.It is notable that the performance of the antennas slightly varies with the specifc setup and the VNA used.Despite the variations, all antennas perform well within the desired frequency range in both confgurations.3 presents the comprehensive test program for the fatigue test, outlining the specifc test parameters for a total of 72,000 cycles.Sinusoidal forces are used according to Table 3, whereas the forces of the two vertical cylinders are inversely phased to generate the targeted bending moment.Te load is expressed as percentage relative to the maximum potential vertical force of 400 kN and the maximum potential horizontal force of 10 kN.

Description of the Measurement Protocol. Table
After several months of curing, initial reference measurements were taken with the grout in dry condition.On November 09 at 10:30 UTC, the water was flled to 450 mm, i.e., 300 mm above the bottom end of the grout.All tests presented below were conducted with the bottom area of the grouted connection in direct contact with water.Tis approach aims to replicate ofshore conditions and enables the generation of realistic particle washout of the grout material.
To identify changes in the structural behavior of the demonstrator during the fatigue test, dynamic tests using sweep sine and random noise excitation were conducted after each load level using the electrodynamic shaker shown in Figure 4. Te results obtained from the dynamic tests were evaluated using modal analysis procedures to identify the trend of change considering eigenfrequencies, mode shapes, and damping properties.All dynamic tests were performed under identical loading conditions to exclude any other infuences on the structural behavior.

Description of Radar Signal Processing.
In order to analyze each radar measurement, the inverse fast Fourier transform (IFFT) is utilized to convert the measured frequency domain data to the time domain.Te absolute value is then calculated to obtain real values from the complex data.Tese real values form a single column in Figure 8(a).Te columns represent diferent time points captured during the measurement period, using a specifc antenna combination (where Tx2 is a transmitter and Rx2 is a receiver).To condense each column into a scalar metric, the root mean square (RMS) is computed for each signal (see ( 1)) as a widely used damage sensitive metric to identify trends in the radar data [32].
Here, x(t) describes the time-domain radar signal and n is the number of sampling points in that signal.Te development of the RMS values over time is shown in Figure 8(b).To visualize the changes that occur throughout one day, the diferential data can also be obtained by subtracting the frst column in Figure 8(a) from all the other columns.Tis diferential data are plotted in Figure 8(c).On the other hand, Figure 8(d) displays the ratio of the current RMS value of Figure 8(b) to the initial RMS value of the entire measurement period.

Structural Control and Health Monitoring
For the sake of simplicity, we will only consider the values shown in Figure 8(d) as a damage indicator in the radar plots that follow.

Documentation of Damage Evolution.
During the fatigue tests, as the load levels increased, massive water leakage of several liters occurred on the top side of the grout.Tis leakage was accompanied by progressive cracking of the grout material.Due to the loads applied by the hydraulic cylinders the grout was partially crushed into smaller pieces and water was pumped upwards through the cracks causing a wash out of the pulverized grout material.Te washed-out material was collected in containers (see Figure 9) for the top (a) and bottom (b) of the grouted connection after an inspection on December 6.Te variation in appearance may be attributed to (a) being scraped and vacuumed when dry, while (b) was scraped when wet and then dried afterwards.It can be assumed that a total of approx.9 kg of the grout material was washed out after the frst revision (72,000 cycles), which corresponds to approx.1% of the total grout material mass.Figure 11 illustrates the development of visual damage of the grout surface over time on a specifc date (November 22) at two opposite sides of the demonstrator (90 °and 270 °).Immediately after the demonstrator was cured, small radial surface cracks became visible.At 12:09 UTC, a wet circumferential crack appeared at the 270 °position, indicated by a yellow arrow.By 12:14 UTC, water could be seen escaping from one of the existing radial cracks at the 90 °position.
At 12:26 UTC, an increasing amount of water emerged from the cracks, flling the space above the grout until only the water surface was visible.By 15:27 UTC, most of the water had disappeared from the surface, although it remained wet.At 16:18 UTC, the surface had dried up further, and the radar measurement was stopped due to water entering the Ethernet port, which required the replacement of it afterwards.Due to the deposition of pulverized grout sediment, a fresh layer has been formed on the existing grout surface, giving it a smooth appearance.See video clip in the Supplementary Materials (available here) for a time-lapse sequence showing the development of the visible damage on November 22.

Analysis of Testing Machine Parameters.
Te hydraulic cylinders' force and displacement data were recorded in parallel to the SHM sensor system.To calculate the moment for further analysis, the values of one vertical cylinder were subtracted from the other.Tis subtraction helps cleaning the signals by eliminating the efect of the hydraulic cylinder heating up, which causes a drift in the displacement signal.
Next, the measured signal was divided at every second zero crossing point.Figure 12 displays the plot of the diference in force and displacement on the left side for the beginning and end of November 22.Even at this stage, it is noticeable that the displacement signal is not as well aligned with the force signal by the end of the day compared to the beginning.Tis misalignment becomes more apparent when the force/displacement is plotted on the right side for the specifc times.
As explained in [33], an increase in area of the hysteresis loops indicates damage.Te area represents the mechanical work of grinding the grout into smaller pieces and can be determined by calculating the integral over the force/displacement curve (see ( 2)).Structural Control and Health Monitoring where W represents the work done by force F ds is the infnitesimal displacement.Another approach is to calculate the Pearson product-moment correlation coefcient (see ( 3)), which in this case measures the linear relationship between force and displacement.It spans from −1 to 1, where −1 denotes a perfect negative correlation, 0 represents no correlation, and 1 indicates a perfect positive correlation.
where n denotes the sample size, a i and b i are the individual sample points, and a and b are their arithmetic mean.Both methods aid in visualizing the progression of damage in the structure from the perspective of the hydraulic system.Figure 13 illustrates the evolution of work and correlation coefcient for all four days of hydraulic loading.It is important to note that the data can only be compared within each load level and not across diferent levels.Te damage depicted in Figure 11 can also be observed in Figure 13.It is evident that only on the fourth day major damage occurred to the grouted connection because the work and correlation coefcient changed signifcantly during the fatigue test while on the other days no meaningful change was observed.Tis is likely because the initial small cracks in the grout only became gradually larger after a signifcant number of load cycles and then propagated.Initially, the grout and steel structure maintain their full integrity as long as the steel does not reach the plasticized state and the concrete's compressive strength is not locally exceeded.Te only conceivable immediate damage is that the concrete in the tension area initially adhered to the steel and then rapidly separated from it.Concrete has limited resistance to tensile forces but can withstand high compressive loads.

Results of Radar Signal Processing.
Te results of the radar measurements conducted for all combinations of transmitting (Tx) and receiving (Rx) antennas are shown in Figure 14.Te presence of gray signifes instances when laboratory staf was likely present.Te presence of people should be taken into account when analyzing the radar data as their movements and equipment operation could infuence the radar measurements.Tese periods were identifed by analyzing camera images during times when the ceiling    Te radar sensor successfully detected the major damage.It is evident from the fgure that signifcant damage occurred exclusively on November 22. Terefore, the radar observed consistent properties initially, followed by rapid changes on November 22. Figure 15 provides a detailed plot of the radar measurement results for all antenna combinations specifcally on that day.Additionally, one dashed line is incorporated in the fgure to visually indicate the initial appearance of water on the grout surface.Another dashed line marks the point in time after which most of the water had fown back into the cracks.But the surface stayed wet and afterwards gradually dried over time.
Te dielectric properties of the dry grout were examined in [31].During the fatigue test the properties of the grout are not homogeneous, since it partly contains water rising in the grouted connection.Tus, a precise localization of the damage must consider this inhomogeneity.In addition, the electromagnetic waves can propagate over several paths from the transmitter to the damage and back to the receiver,  e.g., on a straight path or around the circumference.Figure 16 showcases a simulation of the electric feld (E-feld) in the grouted connection using the CST Microwave Studio for a single antenna.Te cylindrical shape of the demonstrator was unwrapped to enable simulation and plotting of the data.Te top and bottom sections correspond to diferent frequencies: 100 MHz and 2 GHz.Te simulation provides insight into the distribution and behavior of the E-feld in relation to the grouted connection at these specifc frequencies.Te diferent wavelengths corresponding to the lower and higher frequency can be easily observed.Te simulation results illustrate why a rough estimation of the damage position can be possible.At low frequencies, the electric feld of the transmitting antenna spreads throughout the whole demonstrator, while at higher frequencies, the signal tends to be more concentrated below the antenna.Te values obtained from the last radar measurement conducted on November 22 at 16:18 UTC (see Figure 15) are presented in Figure 17 with an arrow indicating the location of the corresponding transmitting and receiving antenna.Te reduction of the damage indicator after the damage event is location-dependent.Te largest change occurred between Tx2 and Rx2, which was also one of the areas where the fnite element analysis (see Figure 2

Vibration Analysis.
A large number of sensors for vibration and acoustic emission measurement were installed on the laboratory demonstrator from outside and inside.Tese enabled versatile monitoring of the grout and fnally damage detection using global and local methods.In this paper, only a small insight to this feld is provided.A total of 100 measurement channels were recorded by means of two measurement systems: 64 strain gauge channels distributed over the height and circumference of the grouted joint, six Pt100 temperature sensors for temperature measurement and compensation, four triaxial micro-electro-mechanical-system (MEMS) acceleration sensors on diferent levels of the test rig, two uniaxial MEMS acceleration sensors on the shaker, two   Using various evaluation methods, the damage state of the test rig can be continuously monitored.One of the global methods is the operational modal analysis (OMA); some results are presented in the next section.However, even statistical features already allow the damage progress to be identifed.For example, in Figure 18, the normalized RMS values of acceleration in the frequency band from 0.5 to 1.5 Hz show not only a step-like increase in the values due to the step-wise increasing load levels, but also a change during the respective load levels.Tese indicate a change in the structural properties, especially the reduction of structural stifness, caused by the hydraulic excitation and thus a change in the frequency spectrum of the acceleration signals.It is also interesting to note that the properties also change transversely to the load direction and that these can also be measured.Te large short-term change in all the curves on November 22 at about 13:40 UTC results from a short switch-of of the test rig.
Te signifcant damage progress during the fourth load stage can be clearly recognized using both, the more global acceleration measurement and the more local strain measurement.As an example, this is shown in Figure 19 using an indicator based on the combination of transfer functions between multiple strain gauges as well as the skewness for selected strain gauge channels [34].A clear change in the characteristics and indicators can be seen, in particular on November 22 starting from approx.12:00 UTC.Tis signifcant damage progress can also be observed by acoustic emission analysis.Tis can be seen impressively in Figure 20, for example, in the increasing cumulative energies [35] and their rates of change for the individual sensors.Features such as the RA value, which is the ratio of amplitude and rise time (see (4)), show corresponding changes.Te diferent rates of change of the acoustic emission energy for the individual sensor channels are already an indication for local accumulations of the acoustic emission sources in the laboratory structure.

RA
Te visual observations (see previous section) correspond to the fndings from the continuous evaluation of the (local) vibration signals as well as the results of the acoustic emission analysis, so that it is possible to monitor both longterm trends and transient damage events, respectively.

Modal
Analysis.An important and powerful tool for damage detection in the feld of vibration analysis is the OMA.Tere are various well-known techniques, which can be used, each with its own advantages and disadvantages [36].Here, a modifed approach of the stochastic subspace identifcation covariance driven algorithm (SSI-COV OMA) was used to analyze the acceleration data [37].Te acceleration was recorded by MEMS sensors during the structure was excited by an electrohydraulic shaker.Te excitation procedure was triggered before each hydraulic load level, in order to track the modal quantities of the laboratory  demonstrator.Te frequency range of the random excitation was relatively high (up to 1000 Hz) and could therefore excite as many modes as possible.
Figure 21 shows the PSD from two channels of accelerometer sensors measuring parallel to the shaker excitation.T1 and T2 denote the acceleration sensors and z is the corresponding measuring direction.Te red dashed lines show the frequency range where a dominant mode is sensitive to damage.Tis frequency range has been selected for further analysis.To smooth the results of the PSD, a running mean flter was used in this case for display purposes only.
According to the results of the OMA as given in Table 4, a clear reduction of the eigenfrequency and an increase of damping are a defnite indication for the damage manifested as a decrease of the structural stifness and an increase of the friction in the grout.A degradation of the grouted connection had already occurred after the frst load level.No plastic deformation was observed in the steel components, suggesting that the damage was confned to the grout material itself.Te suspicion was confrmed as the load level was increased, with the appearance of cracks in the grouted joint.

Conclusions
Tis paper presented an innovative approach for in situ monitoring of the grouted connection in ofshore wind turbines using a stepped-frequency continuous wave (SFCW) radar.Te proposed radar-based structural health monitoring system ofers a nondestructive and real-time method for assessing the structural integrity of the grouted connection directly, actively, and automatically.Trough a newly developed large-scale experiment and subsequent fatigue testing, the efectiveness of the radar approach was demonstrated.A rough localization of the damage in relation to the circumference of the grouted connection was possible based on the antenna positions.
For validation, this study also presented visual observations, hydraulic cylinder data, vibration analysis, acoustic emission analysis, and modal analysis.Tese results confrm the radar system's ability to detect damage in the grouted connection and efectively monitor the integrity of the turbine foundation.By providing real-time information on the structural health, the proposed approach can enable predictive maintenance strategies and the optimization of the operational lifetime as well as a reduction of maintenance cost of ofshore wind turbines in the future.

Figure 1 :
Figure 1: (a) Location of the grouted connection on a 3.6 MW ofshore wind turbine [5]; (b) illustration of the grouted connection showing the application of the radar antennas on top of the grout.

Figure 2 :
Figure 2: (a) Overview of the laboratory demonstrator with labels of its main components; (b) results of the fnite element analysis showing the one of the areas with the highest stress in red at the 180 °position in the grouted region.

Figure 3 :
Figure 3: (a) Gap between the metal tubes with shear keys that is subsequently flled with grout; (b) radar sensor mounted on the outer metallic tube together with the antennas directly coupled to the grout from the top.Additionally, a hydraulic cylinder and other sensors are shown.

Figure 4 :
Figure 4: Electrodynamic shaker used for dynamic testing during the fatigue test.

Figure 5 :
Figure 5: Radar system with open housing cover.

Figure 6 :
Figure 6: (a) Antenna placement on the grouted connection with an arrow indicating the load direction for bending; (b) two antennas installed on the grouted joint without coaxial cable.

Figure 7 :
Figure7: Input refection coefcient S 11 of all fve antennas used in this study; (a) on a simplifed structure described in[31] (using a Keysight M9807A vector network analyzer) and (b) on the fnal laboratory demonstrator structure (using a NanoVNA V2 Plus4 vector network analyzer).

Figure 8 :
Figure 8: Radar signal processing (yellow indicates missing data): (a) absolute time domain data from one antenna combination (Tx2 as transmitter and Rx2 as receiver); (b) root mean square (RMS) of each column shown in (a); (c) diferential time domain data calculated by subtracting the frst column of all columns shown in (a); (d) ratio of the RMS in (b) compared to the initial RMS value in (b).

Figure 9 :Figure 10 :
Figure 9: Washed-out material (a) on the top and (b) on the bottom of the grouted connection.(a) Being scraped and vacuumed when dry; (b) being scraped while wet and subsequently dried.

Figure 11 :Figure 12 :
Figure 11: Evolution of the visible damage on November 22 at two opposite sides of the demonstrator (90 °and 270 °).Small radial surface cracks are visible immediately after curing, with a wet circumferential crack appearing at the 270 °position by 12:09 UTC (indicated by a yellow arrow).At 12:14 UTC, water starts escaping from one of the radial cracks at the 90 °position.At 12:26 UTC, a growing volume of water surfaced from the cracks, gradually flling the area above the grout, leaving only the water surface visible.Te water surface remains visible until 15:27 UTC when most of it disappears, leaving the surface still wet.At 16:18 UTC, the surface dries up further.

Figure 13 :Figure 14 :
Figure 13: Work (integral of the force/displacement curve) (a) and Pearson product-moment correlation coefcient of force and displacement (b).

Figure 15 :
Figure 15: Results of the radar measurements for all antenna combinations on November 22. Gray indicates the times when stafs were present in the laboratory.Yellow indicates the times of the fatigue test.Te dashed line shows the frst visual appearance of water on the surface of the grout and the disappearance of the water (although the surface was still wet afterwards).

Figure 16 :
Figure 16: Simulation of the E-feld of one antenna placed on the grouted connection at 100 MHz (a) and 2 GHz (b).Te black rectangle shows the boundary of the metal/grout area.

Figure 17 :
Figure 17: Values of the last radar measurement on November 22 at 16:18 UTC in relation to the position of the antenna combination on the grouted connection.

Figure 18 :
Figure 18: Normalized RMS values of acceleration in the frequency band 0.5-1.5 Hz in load direction (a) and transverse to load direction (b) measured by four MEMS acceleration sensors at the measurement positions T1, T2, Gi0-1, and Ga0-1 during hydraulic excitation at four load levels.

Figure 19 :
Figure 19: Transfer function-based damage indicator (a) and skewness of strain signals (b) during hydraulic excitation at four load levels and the measurement positions Ga1-Ga5 at the outer pipe resp.Gi0-Gi5 at the inner pipe.

Figure 20 :
Figure 20: Cumulative energy (a), median fltered rate of change of the cumulative energy (b), and RA value (c) of acoustic emission events.

Figure 21 :
Figure 21: PSD over the whole frequency range.Te red dashed lines show as an example the range of interest for the modal analysis.

Table 1 :
Operating parameters of the radar sensor.

Table 2 :
Antenna placement on the grouted connection.

Table 3 :
Procedure and test parameters for fatigue test.

Table 4 :
Evaluation of the modal analysis of damage mode.