Real-time monitoring system and method for scouring depth of offshore wind power pile foundation

Abstract

The invention provides a real-time monitoring system and method for the scouring depth of an offshore wind power pile foundation, and belongs to the technical field of ocean pile foundation scouring, and the real-time monitoring system and method for the scouring depth of the offshore wind power pile foundation comprise a fan which comprises a wind wheel, a cabin, a tower barrel and a sleeve; the acceleration sensor is mounted at a maintenance platform cross beam at the top of a cabin of the fan and is positioned behind a wind wheel rotating plane; the edge computing terminal is in communication connection with the acceleration sensor, is internally provided with a modal recognition module and is used for processing the vibration signal into the current first-order frequency of the fan and the pile foundation system; the data platform receives and stores the current first-order frequency in a wireless mode and is internally provided with a mapping module, and the mapping module pre-stores the mapping relation between the relative frequency and the scouring depth; the visual client side is connected with the data platform, and the problem that in the prior art, due to the fact that the first-order frequency variation of a pile foundation is extremely small and interference of working conditions is large, the real-time monitoring precision of the scouring depth of the offshore wind power single pile is insufficient is solved.

Description

Technical Field

[0001] This invention belongs to the field of marine pile foundation scour technology, and more specifically, relates to a real-time monitoring system and method for the scour depth of offshore wind power pile foundations. Background Technology

[0002] The global energy structure is undergoing profound adjustments, and offshore wind power, with its abundant resources, proximity to load centers, and large-scale development potential, has become a crucial tool for countries to achieve carbon neutrality goals. In recent years, my country's coastal wind farm construction has accelerated, with monopile foundations becoming the mainstream due to their convenient construction, controllable costs, and flexible adaptability to turbine locations. However, the complex marine environment, with its combined effects of tides, waves, and turbulence, keeps the soil around the piles under high shear for extended periods, making it highly susceptible to localized scour. Once a scour pit forms, the pile embedment length shortens, horizontal bearing stiffness decreases, and the structure's natural frequency drifts. Under extreme conditions, this can induce resonance or fatigue accumulation, directly threatening the safety of the turbine throughout its entire lifecycle. Traditional underwater monitoring methods are significantly affected by external factors and have high maintenance costs, making them unsuitable for dynamic monitoring. While some technologies attempt to invert scour depth through changes in the pile's natural frequency, the complex excitation of offshore wind turbines means that frequency drift often falls to the thousandths or even ten-thousandths of a percent. The accuracy of identification is significantly amplified by the coupling of multiple factors, including signal-to-noise ratio, operating conditions, and sensor drift, resulting in insufficient engineering practicality. Scouring has thus become a common, hidden, and dynamic key risk point in offshore wind power operation and maintenance. There is an urgent need for a monitoring method that is adapted to the marine environment, meets engineering requirements in terms of accuracy, and can be operated in real time to support the transformation from passive maintenance to proactive operation and maintenance. Summary of the Invention

[0003] In view of this, the present invention provides a real-time monitoring system and method for the scour depth of offshore wind power pile foundations, which solves the problem of insufficient real-time monitoring accuracy of the scour depth of single offshore wind power piles due to the extremely small change in the first-order frequency of the pile foundation and the large interference from operating conditions.

[0004] This invention is implemented as follows:

[0005] This invention provides a real-time monitoring system for the scour depth of offshore wind turbine foundation piles, comprising:

[0006] The wind turbine includes a rotor, nacelle, tower, and sleeve. The rotor is rotatably connected to the front of the nacelle via a hub and a main shaft. The nacelle houses the transmission components. The bottom plate of the nacelle is rigidly fixed to the top flange of the tower, making the tower the only force transmission path between the nacelle and the substructure. The tower is a hollow conical steel cylinder, and its bottom flange is bolted to the upper flange of the sleeve to form a transition section. The sleeve is a cylindrical steel casing that is fitted onto the outside of the steel pipe pile and extends downward below the mud surface. The sleeve and the steel pipe pile are connected by grouting or flanges to form an integral pile foundation, thereby providing a fixed support point for the tower.

[0007] An accelerometer is installed on the crossbeam of the maintenance platform on top of the wind turbine nacelle, behind the rotor's rotation plane and above the tower's centerline, to continuously acquire vibration signals from the upper structure of the wind turbine.

[0008] The edge computing terminal communicates with the acceleration sensor and has a built-in modal recognition module to process vibration signals into the current first-order frequency of the wind turbine and pile foundation system.

[0009] The data platform receives and stores the current first-order frequency wirelessly, and has a built-in mapping module. The mapping module pre-stores the mapping relationship between relative frequency and scour depth, which is used to convert the current measured frequency into real-time scour depth.

[0010] A visualization client, connected to the data platform, is used to display the pile foundation scour depth and alarm information to users in real time.

[0011] A triaxial accelerometer is rigidly mounted on the top of the wind turbine nacelle (near the center of gravity and away from the blade rotation plane), with a sampling frequency of 100Hz and a sensitivity of 100mV / g. The sensor is connected to an edge computing terminal (ARM Cortex-A72 quad-core, Ubuntu 20.04 system) inside the nacelle cabinet via shielded twisted-pair cable. The terminal runs a real-time thread: every 10 minutes, it captures 600 seconds of raw signal, removes the mean and trend, and then calls a random subspace recognition algorithm to output the current first-order frequency. The terminal uploads the frequency value to a cloud database via a 4G / 5G router. A relative frequency-to-scour depth mapping table is pre-stored in the cloud, and the Node-RED streaming computing engine performs table lookups and writes the data to the real-time depth field. The user-end web interface retrieves depth data every 30 seconds; if the limit is exceeded, a red alarm card pops up and a prompt sound plays.

[0012] Based on the above technical solution, the real-time monitoring system for scour depth of offshore wind power pile foundations of the present invention can be further improved as follows:

[0013] The modality recognition module employs a random subspace recognition algorithm and can automatically adjust calculation parameters under different sea conditions. Before each calculation, the edge terminal estimates the signal-to-noise ratio using the Welch method. If the signal-to-noise ratio is less than a predetermined value, it automatically increases the number of rows and blocks in the Hankel matrix, increases the system order, and enables automatic clustering of the stability graph to lock a reliable first-order frequency in a stable graph manner, thereby controlling the frequency recognition error to within five per thousand.

[0014] The modality recognition module employs the Random Subspace Identification (SSI) algorithm, with code based on the Python SSI-cov package. To combat sea state changes, the edge terminal estimates the signal-to-noise ratio (SNR) using the Welch method before each calculation. The specific steps for Welch's SNR estimation are as follows:

[0015] Data segmentation: Take the current 10-minute raw acceleration sequence x(n), sampling frequency fs = 100Hz, a total of 60,000 points; divide it into 117 segments with 1024 points per segment and 50% overlap (the last segment with less than 1024 points is discarded).

[0016] Window function processing: Multiply each segment by a Hamming window to reduce spectral leakage; the window function normalization coefficient G = 1.5868 is used for subsequent power spectrum amplitude correction.

[0017] Periodogram averaging: Perform an FFT on each windowed segment to obtain periodogram I. i (k)=∣FFTx i (n)∣ 2 ;

[0018] Among them, I i (k) is the periodogram (initial power spectral density) of the i-th data segment, FFTx i (n) represents the i-th time-domain signal x i (n) Perform a Fast Fourier Transform.

[0019] The power spectrum estimate Pxx(k) is obtained using the Welch averaging formula:

[0020] Pxx(k)=(1 / (117·G·fs))·ΣI i (k);

[0021] Where G represents the power recovery coefficient of the window function (1.5868 for the Hamming window); fs is the sampling frequency in Hz.

[0022] Signal band and noise band division:

[0023] Signal band B s : 0.15–0.4Hz (including first-order bending of the tower);

[0024] Low-frequency noise band B n 1: 0–0.05Hz (tidal level, long-period surge);

[0025] High-frequency noise band B n 2: 0.8–2Hz (high frequency of blades, mechanical noise).

[0026] SNR calculation:

[0027] signal power

[0028] Noise power P n =∫B n 1Pxx(k)dk+∫B n 2Pxx(k)dk;

[0029] SNR = 10·log10(P) s / P n ).

[0030] The calculation results are written to shared memory for subsequent SSI parameter selection threads to read.

[0031] Automatic parameter adjustment rules:

[0032] SNR≥10dB: Keep the default i=20, N=30;

[0033] 5dB≤SNR<10dB: i=30, N=40;

[0034] SNR < 5dB: i = 40, N = 50, and stable graph clustering (ε) is enabled. f <1%, MAC>0.9); i represents the number of rows and blocks in the Hankel matrix (SSI parameter); N represents the system order (SSI parameter); ε f The frequency stability threshold is set to 1%; the MAC represents the modal confidence criterion, set to >0.9.

[0035] Experimental results show that this strategy reduces the standard deviation of first-order frequency identification from 0.018 Hz to 0.010 Hz, meeting the requirement of "error control within five per thousand".

[0036] Furthermore, the mapping relationship between the relative frequency and the scour depth is obtained by measuring the reference scour depth and simultaneously collecting the reference first-order frequency before the wind turbine is put into operation, and then repeatedly measuring at different scour depths to obtain multiple sets of data points, which are then fitted to form a monotonically decreasing curve for subsequent depth inversion.

[0037] The relative frequency versus scour depth calibration curve of the mapping module is obtained in the following way:

[0038] After the wind turbine is installed and before it is put into operation, the reference scour depth is obtained by sonar or magnetic measurement, and the reference first-order frequency is recorded and the relative frequency is calculated. The above steps are repeated at different scour depths to obtain discrete points, and the calibration curve with monotonically decreasing is obtained by fitting with the least squares method.

[0039] The mapping relationship between relative frequency and scour depth was completed two weeks before the wind turbine was put into operation.

[0040] A multi-functional scour and siltation survey vessel was hired, and a dual-frequency sonar (400kHz / 900kHz) was used to scan around the pile 360° to output a digital ground model around the pile. The difference in elevation between the lowest point of the scour pit and the mud surface was taken as the reference depth.

[0041] Acceleration data was collected synchronously 24 hours a day, and the first-order frequency was calculated every 4 hours. The average value was taken as the reference frequency.

[0042] Subsequently, taking advantage of the natural scouring differences during the spring tide, the above-mentioned combined sonar and acceleration measurements were repeated at eight depth levels: 0m, 1m, 2m...8m, to obtain eight sets of (depth, frequency) points.

[0043] A monotonically decreasing curve is obtained by fitting with cubic splines and written into the cloud mapping module.

[0044] Furthermore, a self-calibration unit is also provided to re-acquire the reference first-order frequency during the annual low wind and wave window and compare it with the original reference frequency. If the deviation exceeds one percent, the mapping relationship between the relative frequency and the scour depth is automatically updated, and the mapping curve is re-normalized to eliminate errors caused by sensor drift and structural aging.

[0045] A self-calibration unit (RTC clock + Shell script) has been added to the engine room cabinet. The low wind and wave window refers to the first occurrence in April each year of wind speed <3m / s, significant wave height <0.5m, and unit shutdown lasting more than 6 hours; the script will automatically trigger during this window.

[0046] With MQTT push disabled, continuously collect acceleration data for 1800 seconds;

[0047] Calculate the new first-order frequency using the same SSI parameters as the initial calibration;

[0048] If the deviation from the original reference frequency exceeds 1%, the old reference frequency is replaced with the new frequency, and the mapping curve is renormalized (the cloud-based Node-RED automatically performs CSV rewriting), thereby eliminating sensor zero drift and structural stiffness degradation errors within one year.

[0049] This invention provides a method for real-time monitoring of the scour depth of offshore wind turbine foundations, comprising the aforementioned real-time monitoring system for the scour depth of offshore wind turbine foundations, and the method comprising the following steps:

[0050] S10: Install acceleration sensors on the top of the wind turbine nacelle, establish a coupled dynamic model of the wind turbine and pile foundation under different scouring conditions, and calculate the first-order frequency samples corresponding to different scouring depths.

[0051] S20: Before the wind turbine is put into operation, the current scour depth is measured as the reference depth, and acceleration data is collected to extract the reference first-order frequency;

[0052] S30: Using the reference depth and reference first-order frequency as the basis for normalization, and combining the model samples to construct the mapping curve between relative frequency and scour depth;

[0053] S40: During wind turbine operation, real-time acceleration data is collected and the current first-order frequency is identified. The current scour depth is obtained by inversion based on the mapping curve.

[0054] S50: When the scouring depth exceeds the set threshold, an alarm is triggered and the data is uploaded to the operation and maintenance platform.

[0055] The steps are as follows:

[0056] Modeling: A coupled dynamic model of the tower, pile foundation, and seabed was established in Abaqus software. The soil around the piles was selected as a single layer, divided into an outer soil portion, and the py curve was used. The scour pits around the pile foundation could be considered as approximately inverted cones. The bottom elevation of the pits was gradually reduced, and first-order modal frequencies were extracted in batches to form a sample library. The py curve is an experimental curve in soil mechanics describing the nonlinear relationship between the lateral resistance p of the pile side soil and the lateral displacement y of the pile, used to simulate the horizontal elastic-plastic constraint of the pile surrounding soil on the tower-pile foundation system. The frictional contact between the soil and the pile foundation was set using the ODB import method to realize the soil spring effect in the coupled model of the tower, pile foundation, and seabed.

[0057] Reference acquisition: Complete the measurement of reference depth and reference frequency;

[0058] Curve construction: Normalize all frequencies in the sample library to the reference frequency to obtain a relative frequency sequence, and import it into MATLAB along with the corresponding depth for cubic spline fitting to generate a mapping curve;

[0059] Real-time inversion: When the wind turbine is running, the edge terminal outputs the current first-order frequency every 10 minutes. The cloud converts it into a relative frequency and then looks up the depth on the curve. If the depth is greater than the design threshold (e.g., 0.7D), an alarm is immediately pushed to the operation and maintenance platform via HTTPS POST.

[0060] Based on the above technical solution, the real-time monitoring method for scour depth of offshore wind power pile foundations of the present invention can be further improved as follows:

[0061] Furthermore, before identifying the first-order frequency in S40, the acceleration data is first bandpass filtered to remove interference frequencies from tide levels, ships, and blades, thereby improving the accuracy of identification.

[0062] The specific steps for bandpass filtering to remove interference frequencies are as follows:

[0063] Interference source spectrum calibration:

[0064] Raw acceleration data was continuously collected for 24 hours under the "background conditions" of wind turbine shutdown, no wind, and no waves. The background power spectrum was obtained by Welch method, confirming that: tidal level and long-period swell energy are concentrated in 0–0.05Hz; ship impact and low-frequency mooring sway are in 0.05–0.08Hz; the blade 1P passing frequency range is 0.15–0.35Hz (varying with rotational speed); the blade 3P frequency range is 0.45–1.05Hz; and the first-order bending mode of the tower is located in 0.28–0.32Hz (target signal).

[0065] Filter design:

[0066] An 8th-order zero-phase Butterworth IIR bandpass filter is used, with a passband of 0.15–0.4Hz: the lower cutoff of 0.15Hz can suppress tide levels, ships, and 1P low frequencies; the upper cutoff of 0.4Hz can suppress 3P and higher high-frequency components; zero-phase is achieved using Matlab's `filtfilt` or Python's `scipy.signal.filtfilt` to ensure no phase distortion.

[0067] Real-time landing:

[0068] Before each SSI calculation, the edge computing terminal first performs the above filtering on the original 600s signal, and the filtered signal is used for modality recognition.

[0069] Furthermore, to reduce frequency fluctuations caused by differences in operating conditions, multiple sets of acceleration data were collected at low, medium, and high power ranges at the same scouring depth. The average value of the first-order frequency was taken as the representative value of the scouring depth to construct a mapping curve. The low, medium, and high power ranges correspond to three operating segments: 0–30%, 30–70%, and 70–100% of the rated power, respectively.

[0070] Under the same scour depth level (e.g., 2m), three operating segments were selected: 0–30%, 30–70%, and 70–100% of rated power. For each segment, 30 sets of data were continuously collected for 600 seconds, for a total of 90 sets. After offline calculation by the edge terminal, the average first-order frequency was taken as the representative value for that depth, and then the data was fitted to the mapping curve, which significantly suppressed the frequency dispersion caused by differences in operating conditions.

[0071] Furthermore, the steps for constructing the mapping curve in S30 are as follows:

[0072] Time-domain acceleration data at the top of the wind turbine tower were collected under different operating conditions of the coupled dynamic model.

[0073] Modal analysis of the acceleration time-domain data was performed using the random subspace identification method to convert it into the vibration frequency of the pile foundation.

[0074] Before monitoring, the scouring depth of the fan was measured and vibration data of the upper structure of the fan were collected to determine the reference depth and the reference first-order frequency.

[0075] The relative frequencies of the pile foundation structure at different scour depths are calculated and correlated one-to-one with the depth of the pile foundation scour pit. The position of each data point is determined and finally fitted into a mapping curve. The mapping curve is constructed using a piecewise interpolation method.

[0076] The specific steps of piecewise interpolation of the mapping curve are as follows:

[0077] Depth interval division:

[0078] Using pile diameter D as the normalized unit, the scour depth of 0–1.5D is divided into: shallow scour section: 0–0.5D, node step size 0.1D; medium scour section: 0.5–1.0D, node step size 0.2D; deep scour section: 1.0–1.5D, node step size 0.3D.

[0079] The total number of nodes is 6 + 3 + 3 = 12, which balances high resolution in shallow segments with stability in deep segments.

[0080] Interpolation function selection:

[0081] Each segment uses cubic Hermite interpolation (PCHIP) to ensure that: the curve is monotonically decreasing; the first derivative is continuous to avoid Runge oscillations; and the second derivative is allowed to jump at nodes to reduce the risk of overfitting.

[0082] Fitting and Validation:

[0083] The 12 sets of data points (depth, relative frequency) obtained from the field calibration are imported into the MATLAB pchip function, which outputs the piecewise polynomial coefficients and writes them to the cloud JSON configuration file.

[0084] Cross-validation: Using sonar measurement points (n=20) that were not involved in the fitting, the maximum absolute error of the predicted depth was 4.2 cm, R0. 2 =0.99, which meets the accuracy requirement. R 2 R is the coefficient of determination, used to measure the interpretability of a mapping curve for measured data points. 2 =0.99 indicates that the linear correlation between "predicted depth" and "sonar measured depth" reaches 99%, indicating that the mapping curve has an extremely high goodness of fit.

[0085] Furthermore, it also includes an annual self-calibration procedure: under conditions of low wind and waves and wind turbine shutdown, the reference first-order frequency is re-acquired. If the deviation from the original reference frequency exceeds one percent, the reference value is updated and the mapping curve is corrected simultaneously to ensure that the long-term monitoring error does not exceed five centimeters.

[0086] Wind turbine shutdown: The main controller issues a "Shutdown" status code, the blade pitch is adjusted to 90°, the rotor is locked, and this process lasts for at least 30 minutes.

[0087] The wind and waves are relatively small: wind speed ≤3m / s (10-minute average of anemometer on top of the cabin); significant wave height Hs ≤0.5m (10-minute average of wave radar or buoy data on site); tidal range ≤0.3m, to avoid long-period tidal waves affecting frequency identification.

[0088] When the above three conditions are met simultaneously and last for more than 30 minutes, the edge terminal will automatically display the "Allow Self-Calibration" sign and then start the annual reference frequency acquisition process to ensure that the background excitation is negligible.

[0089] Furthermore, the scour depth data is linked with the main control system of the wind turbine. When the scour depth approaches the design limit, the wind turbine power is automatically reduced or a shutdown is triggered to achieve safety control based on scour risk.

[0090] The specific steps for linking scouring depth with main control are as follows:

[0091] Communication interface:

[0092] The cloud establishes an encrypted channel with the wind turbine's SCADA system via OPC UAClient and writes two control tags:

[0093] “ScourDepth realtime "(Float32, unit m); "ScourLimitRatio (Float32, percentage, design limit = 1.0)". ScourDepth realtime Real-time scour depth, i.e., the latest scour pit depth of the pile foundation obtained from cloud inversion (unit: meters), is continuously refreshed and written to SCADA for main control logic judgment. ScourLimitRatio: Scour limit ratio, i.e., the percentage of "current depth / maximum allowable design depth" (unit: %). When this value is ≥70% or ≥90%, the main control executes power reduction or shutdown commands respectively, realizing safety linkage based on scour risk.

[0094] Threshold logic:

[0095] When ScourDepth realtime ≥0.7×design limit, "PowerDerate" is issued from the cloud. 70 "instruction;

[0096] When ScourDepth realtime If the value is ≥0.9 × design limit, issue an "EmergencyStop" instruction.

[0097] After receiving the instruction, the master controller executes the following in sequence:

[0098] 70% power limit: The torque setpoint is reduced to 70% of the rated torque, and the blade pitch is adjusted to maintain the optimal pitch angle for 70% power.

[0099] Emergency stop: Pitch to 90°, disconnect the main circuit breaker, lock the rotor, and send back the "ScourStop" status code.

[0100] Reset and manual intervention:

[0101] Only after subsequent sonar or diving inspections confirm that the scour pit has been backfilled to <0.7 limit, and after maintenance personnel click "ScourReset" through the HMI, will the main controller allow a restart, thus achieving closed-loop safety control.

[0102] Compared with existing technologies, the beneficial effects of the real-time monitoring system and method for scour depth of offshore wind turbine pile foundations provided by this invention are as follows: This invention transforms scour depth assessment into a pile foundation vibration mode identification problem, utilizing the wind turbine's own environmental load as a continuous excitation source, eliminating the need for additional vibration sources or underwater operations, significantly reducing monitoring costs and safety risks. By introducing the concept of relative frequency, the measured results are normalized using a reference frequency, amplifying the numerical differences between different scour states and weakening common-mode interference such as sensor drift, temperature changes, and sea state fluctuations, making minute frequency changes more prominent. The system can achieve engineering-usable resolution with constant scale conversion accuracy. The random subspace identification method replaces traditional modal analysis, stably extracting the first-order bending frequency in marine environments with multiple excitation superpositions and drastic signal-to-noise ratio changes; combined with adaptive parameter adjustment and stability diagram criteria, it avoids mode omissions or false peaks, ensuring long-term consistency. The annual self-calibration strategy automatically updates the reference during low wind and wave windows, eliminating slowly accumulated errors such as structural aging, loose joints, and sensor zero drift, ensuring that the error throughout the entire life cycle remains within the engineering allowable range. Upon exceeding the depth limit, the system can automatically reduce power or shut down the turbine in conjunction with the main control unit, preventing the risk of resonance from escalating and achieving a closed-loop monitoring and control system. The entire method uses a single sensor on the top of the nacelle to collect data, with on-site analysis at the edge terminal and wireless data transmission. It does not alter the turbine structure, does not require underwater operations, and installation and maintenance can be completed within the conventional operation and maintenance window, demonstrating good engineering adaptability and promising prospects for widespread application. Attached Figure Description

[0103] Figure 1 This is a diagram of the coupled dynamic model of the wind turbine structure foundation and the seabed described in this invention.

[0104] Figure 2 This is a structural diagram of a real-time monitoring system for the scour depth of offshore wind power pile foundations according to the present invention.

[0105] Figure 3 Stability graph for SSI algorithm mode identification;

[0106] Figure 4 A flowchart of a real-time monitoring system for the scour depth of offshore wind turbine foundations;

[0107] Figure 5 A flowchart of a real-time monitoring method for the scour depth of offshore wind turbine foundation piles;

[0108] The attached diagram lists the components represented by each number as follows:

[0109] 1. Nacelle; 2. Wind turbine; 3. Tower; 4. Sleeve. Detailed Implementation

[0110] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings.

[0111] like Figures 1-2 The diagram shown is an example of a real-time monitoring system for the scour depth of offshore wind turbine foundations provided by this invention, including:

[0112] The wind turbine includes a rotor 2, a nacelle 1, a tower 3, and a sleeve 4. The rotor 2 is rotatably connected to the front end of the nacelle 1 via a hub and a main shaft. The nacelle 1 houses the transmission components. The bottom plate of the nacelle 1 is rigidly fixed to the top flange of the tower 3, making the tower 3 the only force transmission path between the nacelle 1 and the substructure. The tower 3 is a hollow conical steel cylinder. Its bottom flange is bolted to the upper flange of the sleeve 4 to form a transition section. The sleeve 4 is a cylindrical steel casing that is fitted onto the outside of the steel pipe pile and extends downward below the mud surface. The sleeve 4 and the steel pipe pile are connected by grouting or flanges to form an integral pile foundation, thereby providing a fixed support point for the tower 3.

[0113] An accelerometer is installed on the crossbeam of the maintenance platform at the top of the nacelle 1 of the wind turbine, located behind the rotating plane of the rotor 2 and above the centerline of the tower 3, to continuously acquire vibration signals of the upper structure of the wind turbine.

[0114] The edge computing terminal communicates with the acceleration sensor and has a built-in modal recognition module to process vibration signals into the current first-order frequency of the wind turbine and pile foundation system.

[0115] The data platform receives and stores the current first-order frequency wirelessly, and has a built-in mapping module. The mapping module pre-stores the mapping relationship between relative frequency and scour depth, which is used to convert the current measured frequency into real-time scour depth.

[0116] A visualization client, connected to the data platform, is used to display the pile foundation scour depth and alarm information to users in real time.

[0117] The following describes the specific implementation method of the real-time monitoring system for the scour depth of offshore wind turbine pile foundations, including the turbine structure and its connection with other components.

[0118] I. Overall Layout of Wind Turbine: The wind turbine adopts the form of a single-pile foundation offshore wind turbine generator set. The rotor, nacelle, tower and sleeve are arranged vertically from top to bottom to form an integrated structure for rotation, force transmission and embedding. The overall mass of the wind turbine is 367.6t.

[0119] Wind turbine: Three-bladed horizontal shaft configuration, with the hub rotatably connected to the front of the nacelle via the main shaft, used to capture wind energy and convert it into mechanical torque.

[0120] The nacelle is a rectangular welded steel plate shell that houses the gearbox, generator, lubrication and cooling system. The bottom plate is a ring-shaped steel plate that is connected to the flange at the top of the tower with the same diameter and is connected by high-strength bolts to form a rigid force transmission node, making the tower the only force transmission path between the nacelle and the substructure.

[0121] Tower: A hollow conical steel cylinder with an upper outer diameter that matches the bottom plate of the engine room and an enlarged lower outer diameter. The bottom is equipped with a flange for bolting to the upper flange of the sleeve to form a transition section, so as to achieve smooth load transfer. The height is 107m and the steel pipe thickness is 0.08m.

[0122] Sleeve: A cylindrical steel casing is fitted on the outside of the steel pipe pile and extends downward below the mud surface. The sleeve and the steel pipe pile are connected by grouting or flange to form an integral pile foundation, providing a fixed support for the tower to resist horizontal loads and overturning moments. The upper part of the foundation is 18m high, the lower part of the foundation is buried at a depth of 60m, and the scour angle is 30°.

[0123] II. Accelerometer Sensor Arrangement: The accelerometer sensor is installed on the crossbeam of the maintenance platform at the top of the nacelle, behind the rotor's rotation plane and above the tower's centerline, avoiding the blade sweep vibration source to ensure that the collected vibration signal effectively reflects the overall bending mode of the tower and pile foundation. The sensor is rigidly connected to the crossbeam with bolts, and the signal cable is led down along the C-shaped groove on the inner wall of the nacelle to the edge computing terminal.

[0124] III. Edge Computing Terminal Deployment: The edge computing terminal is fixed next to the electrical cabinet at the bottom of the nacelle. It communicates with the acceleration sensor via a shielded cable and has a built-in modal recognition module for real-time processing of vibration signals and outputting the first-order frequency of the wind turbine and pile foundation system. The terminal shell is connected to the nacelle grounding grid at a single point to meet lightning protection and electromagnetic compatibility requirements.

[0125] IV. Data Platform and Visualization Client Deployment: Edge computing terminals upload first-order frequencies to the remote data platform via 4G / 5G wireless. The data platform has a built-in mapping module that pre-stores the mapping relationship between relative frequencies and scour depths. After completing the depth inversion, the data is pushed to the visualization client so that users can view the scour depth and alarm information in real time.

[0126] The following is a supplementary description of the connection between hardware and software in the real-time monitoring system for the scour depth of offshore wind power pile foundations of this invention.

[0127] I. Hardware Layer Connection Relationships:

[0128] Accelerometer and edge computing terminal: Connected using a low-noise coaxial cable (IEPE standard). The sensor outputs an analog voltage signal, and the terminal is powered by a built-in 24V constant current source. The sampling frequency is 100Hz, and the resolution is 24-bit. The cable is fixed along a C-shaped groove on the inner wall of the cabin, with a bending radius ≥50mm. An outer layer of metal corrugated pipe shielding is added, and it is connected to the cabin grounding grid at a single point to suppress electromagnetic interference.

[0129] Edge computing terminal and cabin PLC: Connected via RS-485 (Modbus-RTU) bus. The terminal acts as the master station, reading wind speed, power, and operating status codes in real time for sea state assessment and SNR estimation. The bus uses twisted-pair shielded cable with a baud rate of 115200bps. The terminal side is opto-isolated and meets IEC 61000-4-5 surge level III.

[0130] Edge computing terminal and remote data platform: Wireless link: The terminal has a built-in 4G / 5G module;

[0131] Message format: JSON compressed, single packet ≤ 1kB, publication cycle 10 minutes, immediate QoS=1 retransmission triggered in case of an anomaly; QoS is the Quality of Service level in the MQTT protocol. QoS=1 means "at least once" delivery: after the message reaches the receiver, an acknowledgment must be sent back. If the sender does not receive an acknowledgment within the specified time, it will automatically retransmit until an acknowledgment is received, thereby ensuring that critical data (such as flushing depth exceeding the limit alarm) is not lost due to network interruptions.

[0132] Disconnection caching: The terminal eMMC reserves a 512MB circular buffer, which is automatically retransmitted after communication is restored to ensure data integrity.

[0133] Data platform and visualization client:

[0134] The platform provides a dual-channel solution: an HTTPS RESTful API for initial client retrieval of historical data, and WebSocket for real-time push notifications of depth, alerts, and curve streams. Push latency is less than 1 second, and it supports simultaneous online access on PC browsers and mobile apps.

[0135] II. Hardware and software co-operation timing example (normal operation):

[0136] In order, they are: sensor output analog signal, edge ADC sampling, real-time filtering + SSI, MQTT publish frequency, cloud node, RED parsing and mapping table lookup, InfluxDB storage depth, WebSocket push, and client real-time refresh curves and alarms.

[0137] In the above technical solution, the modality recognition module adopts a random subspace recognition algorithm and can automatically adjust the calculation parameters under different sea conditions. Before each calculation, the edge terminal estimates the signal-to-noise ratio using the Welch method. If the signal-to-noise ratio is less than the predetermined value, it automatically increases the number of rows and blocks in the Hankel matrix, increases the system order, and enables automatic clustering of the stability graph to lock a reliable first-order frequency in a stable graph manner, so that the frequency recognition error is controlled within five per thousand.

[0138] Furthermore, in the above technical solution, the mapping relationship between relative frequency and scour depth is obtained by measuring the reference scour depth and simultaneously collecting the reference first-order frequency before the wind turbine is put into operation, and then repeatedly measuring at different scour depths to obtain multiple sets of data points, which are then fitted to form a monotonically decreasing curve for subsequent depth inversion.

[0139] Furthermore, the above technical solution also includes a self-calibration unit, which is used to re-acquire the reference first-order frequency during the annual low wind and wave window and compare it with the original reference frequency. If the deviation exceeds one percent, the mapping relationship between the relative frequency and the scour depth is automatically updated, and the mapping curve is re-normalized to eliminate errors caused by sensor drift and structural aging.

[0140] This invention provides a real-time monitoring method for the scour depth of offshore wind turbine pile foundations, comprising the aforementioned real-time monitoring system for the scour depth of offshore wind turbine pile foundations, and the method comprising the following steps:

[0141] S10: Install an acceleration sensor on the top of the wind turbine nacelle 1, establish a coupled dynamic model of the wind turbine and the pile foundation under different scouring conditions, and calculate the first-order frequency samples corresponding to different scouring depths.

[0142] S20: Before the wind turbine is put into operation, the current scour depth is measured as the reference depth, and acceleration data is collected to extract the reference first-order frequency;

[0143] S30: Using the reference depth and reference first-order frequency as the basis for normalization, and combining the model samples to construct the mapping curve between relative frequency and scour depth;

[0144] S40: During wind turbine operation, real-time acceleration data is collected and the current first-order frequency is identified. The current scour depth is obtained by inversion based on the mapping curve.

[0145] S50: When the scouring depth exceeds the set threshold, an alarm is triggered and the data is uploaded to the operation and maintenance platform.

[0146] The following is a detailed description of the specific implementation method for real-time monitoring of scour depth of offshore wind turbine foundations, such as... Figure 4 , Figure 5 As shown.

[0147] Step 1: Establish a coupled dynamic model and obtain first-order frequency samples:

[0148] A triaxial accelerometer is rigidly installed on the crossbeam of the maintenance platform on the top of the cabin.

[0149] A coupled dynamic model of "wind-wave-operating load-wind turbine-tower-pile foundation-seabed" was established using Abaqus software; the soil around the pile was represented by a py curve, and the scour pit was represented by a parametric shell, with the depth di increasing gradually from 0.1 to 1.3D in 0.2D increments (D is the pile diameter);

[0150] The steps for establishing the coupled dynamic model of "wind-wave-operating load-wind turbine-tower-pile foundation-seabed" are as follows:

[0151] In the general finite element platform, three-dimensional solids of the nacelle, tower cylinder, transition section, and steel pipe pile are sequentially built from the top of the tower to below the mud surface; a cylindrical soil domain is generated around it, with a diameter and depth of about ten and fifteen times the pile diameter, respectively, to ensure that the boundary does not significantly interfere with the overall bending mode.

[0152] A deformable cavity with a diameter of approximately five times the pile diameter was created in the outer shell area around the pile, and the depth direction was set to an adjustable parameter of 0.2 times the pile diameter. The bottom elevation of the pit was gradually reduced through the parameter batch processing function, providing a geometric basis for subsequent multi-depth sample calculations.

[0153] The tower, transition section, and steel pipe piles are assigned properties as linear elastic steel. For the soil, a single-layer soil is selected as the modeling object, and the elastic modulus, Poisson's ratio, and density of the surface soil are provided in the survey report. Considering that the wind turbine pile foundation also contains soil, the soil modeling is divided into an outer soil part and an inner soil part, and the Coulomb-Mohr model is selected as the soil constitutive model.

[0154] The seabed surface is fixedly constrained, with only vertical displacement released at the lateral boundaries to avoid additional stiffness to the pile body bending. The nacelle mass, rotor mass, and moment of inertia are applied at the tower top, expressed as concentrated mass points. Wave loads were calculated using Matlab, based on the improved Jonswap spectrum and Morison equations suggested by Heda, to calculate the random wave loads on the wind turbine at cut-in, rated, and cut-out wind speeds. Wind loads were calculated based on time-domain data of random wind loads at various heights under different operating conditions of each wind turbine generated using the Daveport spectrum, yielding the random wind loads on the wind turbine pile foundation, tower, and wind turbine at cut-in, rated, and cut-out wind speeds. Finally, based on the mud surface bending moment calculation formulas proposed by Arany et al. for wind turbine operation 1P and 3P loads, the wind turbine's downwind load caused by 1P and 3P loads was calculated, forming a combined "wind-wave-operation" excitation.

[0155] By setting the scour depth as a variable, the soil elements in the pit are removed sequentially using the parametric scanning function, and the corresponding spring stiffness is simultaneously emptied. Modal analysis is then submitted. The first-order lateral bending frequency at the top of the tower at each depth is extracted to obtain a discrete sample library of relative frequencies and scour depths, which can be used for subsequent normalization and mapping curve fitting.

[0156] For each di, rated wind load + typical wave spectrum (JONSWAP, Hs = 1.5m, Tp = 6s) + three operating power ranges of excitation are applied to extract the first-order lateral bending frequency at the top of the tower in batches, forming a "di-fi,model" sample library for subsequent normalization. di represents the depth of the i-th stage scour pit, Hs is the significant wave height, and Tp is the spectral peak period.

[0157] Step 2: Measure the reference depth and reference frequency before commissioning.

[0158] Before the wind turbine is connected to the grid, a multi-functional survey vessel is hired to use dual-frequency sonar (400 / 900kHz) to scan around the pile 360° and obtain the pile perimeter depth (DTM). The difference in elevation between the bottom of the pit and the mud surface is used as the reference depth d0.

[0159] like Figure 3 As shown, 10 minutes of acceleration time-domain data were collected synchronously. After bandpass filtering, the reference first-order frequency f0 was obtained by random subspace identification (SSI). d0 and f0 were written together into the edge terminal's JSON configuration file as the reference for subsequent normalization.

[0160] Step 3: Construct a relative frequency versus scour depth mapping curve

[0161] Divide fi,model for each di in the sample library by f0 to obtain the relative frequency. Where fi,model corresponds to the first-order lateral bending frequency of the tower top of di, f0 represents the reference first-order frequency measured on site, and Rfi represents the relative frequency (dimensionless, %), which is used for subsequent normalization and mapping curve construction.

[0162] Piecewise interpolation is introduced to perform cubic Hermite fitting on discrete points (di, Rfi) to ensure that the curve is monotonic and the first derivative is continuous; the fitting coefficients are uploaded to the cloud mapping module to form a JSON coefficient table that can be looked up online.

[0163] Step 4: Identify the current first-order frequency and invert the depth online.

[0164] When the wind turbine is running normally, the edge terminal captures 600 seconds of acceleration data every 10 minutes, performs bandpass filtering, SSI identification, and outputs the current first-order frequency f1.

[0165] Calculate relative frequency The current scour depth d1 is obtained by calling the cloud mapping module to look up the table; the table lookup uses a binary search method and takes less than 20ms.

[0166] The results are uploaded with the MQTT message and cached locally, so that the transmission can resume when the network is disconnected.

[0167] Step 5: Over-limit alarm and operation and maintenance linkage

[0168] Node-RED in the cloud continuously compares d1 with design limits:

[0169] ≥0.7D: Issue "PowerDerate" 70 "Instruction; PowerDerate" 70 This indicates a power limiting command, which causes the main control unit of the wind turbine to reduce the real-time active power output to 70% of the rated power, serving as an active load reduction protection when the scour depth approaches the warning value.

[0170] ≥0.9D: Issue an “EmergencyStop” command and push it to the user; EmergencyStop indicates an emergency shutdown command, which causes the wind turbine to immediately switch to the feather position, disconnect the main circuit breaker, lock the rotor, and mark the shutdown reason as “scour over-limit” on the SCADA interface to prevent the risk of resonance or overturning from expanding.

[0171] Operations and maintenance personnel can view in-depth time series curves, download CSV reports, and fill in "review records" to turn off alarms in the visualization client.

[0172] Step 6: Bandpass filtering to remove interference

[0173] Before SSI identification, the original acceleration sequence was subjected to an 8th-order zero-phase Butterworth bandpass filter with a passband of 0.15-0.4Hz, which effectively eliminated the low frequencies of tidal level, ship, and blade 3P. Field calibration showed that the standard deviation of the first-order frequency was reduced by about 42% after filtering.

[0174] Step 7: Averaging across multiple power segments to reduce operational variance

[0175] At the same scour depth level (obtained by sonar or natural scour differences), acceleration data were collected in three segments: 0-30%, 30-70%, and 70-100% of rated power. Each segment had ≥30 samples. The average first-order frequency of the three segments was taken as the representative value for that depth, and then the data were fitted to the S30 curve, which significantly suppressed the frequency dispersion caused by the operating state.

[0176] Step 8: Construct the mapping curve using piecewise interpolation

[0177] The range 0-1.5D is divided into three segments: shallow (0-0.5D, step size 0.1D), medium (0.5-1.0D, step size 0.2D), and deep (1.0-1.5D, step size 0.3D). Cubic Hermite interpolation (PCHIP) is used for each segment to ensure monotonicity and low overfitting. The cross-validation error is ≤4cm.

[0178] Step 9: Annual self-calibration to correct long-term drift

[0179] The annual low wind and wave window (wind speed ≤3m / s, wave height ≤0.5m, wind turbine shutdown ≥30min) is automatically triggered by the self-calibration unit;

[0180] Acquire 1800s of acceleration data and obtain a new reference frequency f0′ using SSI;

[0181] like Then f0 ′ Replace the old baseline and renormalize the mapping curve, while generating an audit log;

[0182] If the difference is greater than 1% for three consecutive times, the system will report "abnormal structural stiffness" to remind manual verification, ensuring that the monitoring error over 20 years is ≤5cm.

[0183] Step 10: Implement risk and security control through master control linkage.

[0184] The cloud establishes an encrypted channel with the wind turbine's SCADA system via the OPC UA Client:

[0185] When d1 ≥ 0.7D, write "PowerDerate". 70 The label indicates that the main controller will automatically reduce power to 70% of the rated power.

[0186] When d1≥0.9D, write the “EmergencyStop” label, and the main controller will execute pitch adjustment to 90°, disconnect the main circuit breaker, and lock the wind turbine.

[0187] Only after subsequent checks confirm that the flushing has been backfilled and "ScourReset" is manually clicked will the main controller allow a reset and restart, thus achieving closed-loop safety control.

[0188] Furthermore, in the above technical solution, before the S40 identifies the first-order frequency, the acceleration data is first bandpass filtered to remove interference frequencies from tide levels, ships, and blades, thereby improving the accuracy of identification.

[0189] Furthermore, in the above technical solution, in order to reduce frequency fluctuations caused by differences in operating conditions, multiple sets of acceleration data are collected at the low, medium and high power ranges at the same scouring depth. The average value of the first-order frequency is taken as the representative value of the scouring depth to construct a mapping curve. The low, medium and high power ranges correspond to three operating ranges of 0–30%, 30–70%, and 70–100% of the rated power, respectively.

[0190] Furthermore, in the above technical solution, the steps for constructing the mapping curve of S30 are as follows:

[0191] Time-domain acceleration data at the top of the wind turbine tower were collected under different operating conditions of the coupled dynamic model.

[0192] Modal analysis of acceleration time-domain data was performed using the random subspace identification method to convert it into the vibration frequency of the pile foundation;

[0193] Before monitoring, the scouring depth of the fan was measured and vibration data of the upper structure of the fan were collected to determine the reference depth and the reference first-order frequency.

[0194] The relative frequencies of the pile foundation structure at different scour depths were calculated and correlated one-to-one with the depth of the pile foundation scour pit. The location of each data point was determined and finally fitted into a mapping curve. The mapping curve was constructed using a piecewise interpolation method.

[0195] Furthermore, the above technical solution also includes an annual self-calibration step: under conditions of low wind and waves and wind turbine shutdown, the reference first-order frequency is re-acquired. If the deviation from the original reference frequency exceeds one percent, the reference value is updated and the mapping curve is corrected synchronously to ensure that the long-term monitoring error does not exceed five centimeters.

[0196] Furthermore, in the above technical solution, the scour depth data is linked with the main control system of the wind turbine. When the scour depth approaches the design limit, the wind turbine power is automatically reduced or a shutdown is triggered, thereby achieving safety control based on scour risk.

[0197] Specifically, the principle of this invention is as follows: The system utilizes ocean wind, waves, and the operating load of the generator unit as continuous broadband excitation to induce minute vibrations in the tower-pile foundation system. An accelerometer located at the top of the nacelle picks up the lateral vibration signal at the top of the tower. After anti-aliasing filtering and analog-to-digital conversion, the signal enters the edge computing terminal. The terminal's internal algorithm first estimates the signal-to-noise ratio using the Welch method, adaptively adjusts the number of rows and blocks i of the Hankel matrix identified in the random subspace, and the system order N, generates a stability graph, and locks in the most reliable first-order bending frequency. The system order N should be equal to the number of non-zero singular values ​​in the matrix. For practical engineering, to improve the system's safety factor, a larger system order can be directly set. This frequency monotonically decreases with increasing scour depth, but the absolute change is extremely small; direct inversion would amplify the error. Therefore, the system uses the reference frequency measured during initial commissioning as a reference to calculate the real-time relative frequency, amplifying the differences between different scour states to the percentile level, and then calls the pre-stored mapping curve to complete the depth inversion. The mapping curve is jointly constructed from coupled dynamic model samples and field calibration data, using piecewise interpolation to ensure high sensitivity for shallow scour and high robustness for deep scour. Annual self-calibration re-measures the reference frequency under low wind and wave conditions and during shutdown. If the deviation exceeds the threshold, the mapping coefficients are automatically updated to eliminate long-term drift. Depth results are transmitted to the cloud via a wireless link, sharing a data bus with the wind turbine's main control system. When the depth approaches the design limit, the system automatically limits power or triggers shutdown, forming a complete closed loop of "excitation-sensing-identification-normalization-inversion-control," achieving proactive, quantitative, and real-time management of scour risks.

[0198] To better understand and implement this invention, the following is an embodiment 1 of a specific application scenario:

[0199] This project involves real-time monitoring of scour depth in monopile foundations of an offshore wind farm located on the southeast coast. The wind farm comprises 50 6MW turbine units, with foundations using 7m diameter steel pipe piles. To monitor the scour around the piles in real time, the project utilizes the real-time monitoring system for scour depth of offshore wind turbine pile foundations based on this invention.

[0200] Accelerometer: A single triaxial IEPE sensor is fixed to the top beam of the nacelle, 0.4m from the tower axis, higher than the rear end of the rotor sweep plane, to avoid direct excitation by the blades; the sensor is connected to the edge computing terminal at the bottom of the nacelle via a shielded cable.

[0201] Edge computing terminal: 316L stainless steel cavity, built-in ARM quad-core motherboard and 4G / 5G module; the terminal is interconnected with the cabin PLC via RS-485 to read wind speed, power and operating status in real time for sea condition judgment.

[0202] Data platform and client: MQTT cluster, InfluxDB and Node-RED are deployed using the existing private cloud in the site; the operation and maintenance duty room and mobile app subscribe to topics simultaneously to realize depth curves, alarm pop-ups and sound and light prompts.

[0203] Before commissioning, the survey vessel uses dual-frequency sonar to scan around the piles to determine the reference depth d0; simultaneously, 10 minutes of acceleration data are collected, and the reference first-order frequency f0 is obtained through bandpass filtering and SSI identification, and both are written into the terminal reference file. A coupled dynamic model is established, and model frequency samples are extracted according to the 0-1.5D thrust gradient; acceleration is collected on-site at different tide levels and power ranges, and the relative frequency is obtained by averaging multiple segments and normalizing, and then fitted with the depth samples to form a piecewise cubic Hermite mapping curve, and the coefficients are uploaded to the cloud. During operation, every 10 minutes, 600 seconds of data are automatically extracted, bandpass filtered, and SSI is used to obtain the current first-order frequency f1, the depth d1 is retrieved from the mapping curve, and reported via MQTT. The cloud continuously compares d1 with the design limit: when ≥0.7D, "PowerDerate" is written to SCADA via OPC UA. 70 The main controller automatically reduces power to 70%; when the error is ≥0.9D, "EmergencyStop" is written and pushed to the user, and the wind turbine immediately stops feathering. During the annual low wind and wave window, the terminal automatically collects the new reference frequency. If the deviation from the original f0 is >1%, the reference is updated and the mapping curve is renormalized to eliminate long-term drift.

[0204] In the two years since the project was put into operation, the system has maintained stable operation during typhoons, high tides, and seasonal dry periods; compared with the annual sonar remeasurement, the maximum absolute error in scour depth is 4.1 cm, R 2 =0.99; Both depth over-limit events successfully triggered power reduction and shutdown, avoiding resonance risk, verifying the reliability, accuracy and engineering applicability of the invention in real marine environments.

[0205] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any changes or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in the present invention should be included within the scope of protection of the present invention.

Claims

1. A real-time monitoring system for the scour depth of offshore wind turbine foundation piles, characterized in that, include: The wind turbine includes a rotor, nacelle, tower, and sleeve. The rotor is rotatably connected to the front of the nacelle via a hub and a main shaft. The nacelle houses the transmission components. The bottom plate of the nacelle is rigidly fixed to the top flange of the tower, making the tower the only force transmission path between the nacelle and the substructure. The tower is a hollow conical steel cylinder, and its bottom flange is bolted to the upper flange of the sleeve to form a transition section. The sleeve is a cylindrical steel casing that is fitted onto the outside of the steel pipe pile and extends downward below the mud surface. The sleeve and the steel pipe pile are connected by grouting or flanges to form an integral pile foundation, thereby providing a fixed support point for the tower. An accelerometer is installed on the crossbeam of the maintenance platform on top of the wind turbine nacelle, behind the rotor's rotation plane and above the tower's centerline, to continuously acquire vibration signals from the upper structure of the wind turbine. The edge computing terminal communicates with the acceleration sensor and has a built-in modal recognition module to process vibration signals into the current first-order frequency of the wind turbine and pile foundation system. The data platform receives and stores the current first-order frequency wirelessly, and has a built-in mapping module. The mapping module pre-stores the mapping relationship between relative frequency and scour depth, which is used to convert the current measured frequency into real-time scour depth. A visualization client, connected to the data platform, is used to display the pile foundation scour depth and alarm information to users in real time.

2. The real-time monitoring system for scour depth of offshore wind turbine foundations according to claim 1, characterized in that, The modality recognition module adopts a random subspace recognition algorithm and can automatically adjust the calculation parameters under different sea conditions. Before each calculation, the edge terminal estimates the signal-to-noise ratio using the Welch method. If the signal-to-noise ratio is less than the predetermined value, it automatically increases the number of rows and blocks in the Hankel matrix, increases the system order, and enables automatic clustering of the stability graph to lock a reliable first-order frequency in a stable graph manner, so that the frequency recognition error is controlled within five per thousand.

3. The real-time monitoring system for scour depth of offshore wind turbine foundations according to claim 2, characterized in that, The mapping relationship between relative frequency and scour depth is obtained by measuring the reference scour depth and simultaneously collecting the reference first-order frequency before the wind turbine is put into operation, and then repeatedly measuring at different scour depths to obtain multiple sets of data points. The data are then fitted to form a monotonically decreasing curve, which is used for subsequent depth inversion.

4. The real-time monitoring system for scour depth of offshore wind turbine foundations according to claim 3, characterized in that, It is also equipped with a self-calibration unit, which is used to re-acquire the reference first-order frequency during the annual low wind and wave window and compare it with the original reference frequency. If the deviation exceeds one percent, the mapping relationship between relative frequency and scour depth is automatically updated and the mapping curve is re-normalized to eliminate errors caused by sensor drift and structural aging.

5. A method for real-time monitoring of scour depth of offshore wind turbine foundation piles, characterized in that, The method includes a real-time monitoring system for the scour depth of offshore wind turbine foundations as described in any one of claims 1-4, the method comprising the following steps: S10: Install acceleration sensors on the top of the wind turbine nacelle, establish a coupled dynamic model of the wind turbine and pile foundation under different scouring conditions, and calculate the first-order frequency samples corresponding to different scouring depths. S20: Before the wind turbine is put into operation, the current scour depth is measured as the reference depth, and acceleration data is collected to extract the reference first-order frequency; S30: Using the reference depth and reference first-order frequency as the basis for normalization, and combining the model samples to construct the mapping curve between relative frequency and scour depth; S40: During wind turbine operation, real-time acceleration data is collected and the current first-order frequency is identified. The current scour depth is obtained by inversion based on the mapping curve. S50: When the scouring depth exceeds the set threshold, an alarm is triggered and the data is uploaded to the operation and maintenance platform.

6. A method for real-time monitoring of scour depth of offshore wind turbine foundations according to claim 5, characterized in that, Before identifying the first-order frequency in S40, the acceleration data is first bandpass filtered to remove interference frequencies from tide levels, ships, and blades, thereby improving the accuracy of identification.

7. A method for real-time monitoring of scour depth of offshore wind turbine foundations according to claim 6, characterized in that, To reduce frequency fluctuations caused by differences in operating conditions, multiple sets of acceleration data were collected at low, medium, and high power ranges at the same scouring depth. The average value of the first-order frequency was taken as the representative value of the scouring depth to construct a mapping curve. The low, medium, and high power ranges correspond to three operating segments: 0–30%, 30–70%, and 70–100% of the rated power, respectively.

8. A method for real-time monitoring of scour depth of offshore wind turbine foundations according to claim 7, characterized in that, The steps for constructing the mapping curve of S30 are as follows: Time-domain acceleration data at the top of the wind turbine tower were collected under different operating conditions of the coupled dynamic model. Modal analysis of the acceleration time-domain data was performed using the random subspace identification method to convert it into the vibration frequency of the pile foundation. Before monitoring, the scouring depth of the fan was measured and vibration data of the upper structure of the fan were collected to determine the reference depth and the reference first-order frequency. The relative frequencies of the pile foundation structure at different scour depths are calculated and correlated one-to-one with the depth of the pile foundation scour pit. The position of each data point is determined and finally fitted into a mapping curve. The mapping curve is constructed using a piecewise interpolation method.

9. A method for real-time monitoring of scour depth of offshore wind turbine foundations according to claim 8, characterized in that, It also includes an annual self-calibration step: under conditions of low wind and waves and wind turbine shutdown, the reference first-order frequency is re-acquired. If the deviation from the original reference frequency exceeds one percent, the reference value is updated and the mapping curve is corrected synchronously to ensure that the long-term monitoring error does not exceed five centimeters.

10. A method for real-time monitoring of scour depth of offshore wind turbine foundations according to claim 9, characterized in that, The scour depth data is linked with the main control system of the wind turbine. When the scour depth approaches the design limit, the wind turbine power is automatically reduced or a shutdown is triggered to achieve safety control based on scour risk.

Images