Multidimensional parameter health monitoring system and method for offshore wind power tower

Abstract

The invention discloses an offshore wind power tower multi-dimensional parameter health monitoring system and method. The system comprises a multi-source sensing module, a data preprocessing module, a feature extraction module, a state evaluation module and an early warning module. The multi-source sensing module integrates a three-axis acceleration sensor, an optical fiber strain gauge, a corrosion rate probe and the like, and collects structural response, environmental load and corrosion data in real time. The data preprocessing module adopts a self-adaptive Kalman filtering and multi-mode standardization method to improve data quality; the feature extraction module extracts 28-dimensional feature vectors through time domain statistics, frequency domain power spectrum and wavelet transform; the state evaluation module outputs a health state index (SHI) based on a bidirectional LSTM-Attention model, and realizes uncertainty quantification in combination with Monte Carlo Dropout; and the early warning output module triggers three-level grading early warning according to an SHI threshold value, and is linked with a digital twin system to realize three-dimensional visualization and intelligent report generation, so that the operation and maintenance intelligent level of the offshore wind power tower can be effectively improved.

Description

Technical Field

[0001] This invention relates to the field of offshore wind turbine tower health monitoring technology, and in particular to a multi-dimensional parameter health monitoring system and method for offshore wind turbine towers. Background Technology

[0002] As the global energy structure shifts towards renewable energy, offshore wind power, as a crucial component of clean energy, has seen its installed capacity grow rapidly. However, offshore wind turbines operate in a complex marine environment, enduring multiple dynamic loads from wind, waves, and currents, while also facing degradation issues such as seawater corrosion and material fatigue, leading to a significant increase in structural safety risks. Traditional wind turbine health monitoring primarily relies on manual inspections and single-parameter monitoring, which suffers from the following technical shortcomings:

[0003] Limited Monitoring Dimensions: Existing technologies often focus on monitoring single physical quantities such as structural vibration or strain, making it difficult to comprehensively reflect the true state of a structure under multi-field coupling. For example, accelerometers alone cannot identify cross-sectional stiffness degradation caused by corrosion, while strain monitoring is insufficiently sensitive to early micro-damage. Furthermore, the lack of correlation analysis between environmental loads (such as extreme wind speeds and wave impacts) and structural response makes damage tracing difficult.

[0004] Insufficient data fusion and processing capabilities: Offshore wind turbines deploy a variety of sensor types (such as electrical, fiber optic, and electrochemical sensors), with a wide range of data sampling frequencies (from 1Hz to 1kHz). Traditional monitoring systems lack the ability to fusion and process heterogeneous data. Existing technologies mostly employ fixed-parameter filtering algorithms, which cannot adapt to the non-stationary noise characteristics of the marine environment (such as the time-varying frequency components of ocean waves), resulting in low signal-to-noise ratios and difficulties in extracting key features.

[0005] Limitations of health assessment methods: Existing assessment models are mostly based on threshold comparisons or shallow machine learning algorithms, relying on human experience to set feature parameters, making it difficult to capture the nonlinear evolution of structural degradation. For example, while traditional LSTM networks can process time-series data, they lack attention mechanisms for key time steps, leading to significant long-term dependency issues. Furthermore, model training relies on large amounts of labeled data, while damage samples from offshore wind turbine towers are scarce, resulting in insufficient transfer learning capabilities.

[0006] The early warning mechanism is lagging: Existing early warning systems mostly use a two-level threshold mechanism, and the channels for pushing early warning information are limited, which cannot meet the needs of tiered response. For example, minor damage only triggers an email notification, while emergency shutdown orders require manual confirmation, which may lead to response delays and catastrophic accidents in extreme weather. In addition, the lack of uncertainty quantification methods results in low confidence of assessment results, which can easily lead to false alarms or missed alarms.

[0007] Insufficient digitalization: Traditional monitoring systems lack 3D visualization and digital twin technology support, requiring maintenance personnel to rely on 2D charts to analyze data, making it difficult to intuitively locate damage. Furthermore, historical data does not form structural health evolution trend maps, failing to provide a basis for predicting remaining lifespan and making it difficult to formulate preventative maintenance strategies.

[0008] Therefore, it is necessary to provide a new multi-dimensional parameter health monitoring system and method for offshore wind turbine towers to solve the above-mentioned technical problems. Summary of the Invention

[0009] To address the aforementioned technical problems, this invention provides a multi-dimensional parameter health monitoring system and method for offshore wind turbine towers.

[0010] The multi-dimensional parameter health monitoring system for offshore wind turbine towers provided by this invention includes:

[0011] Multi-source sensing module for real-time acquisition of multi-dimensional parameters of wind turbine tower structure;

[0012] The data preprocessing module performs filtering and normalization on the collected data;

[0013] The feature extraction module extracts time-domain and frequency-domain features from the preprocessed data;

[0014] The status assessment module evaluates the health status of wind turbine towers based on machine learning models.

[0015] The early warning output module generates a graded early warning signal when the evaluation result exceeds the threshold.

[0016] Preferably, the multi-source sensing module includes:

[0017] The structural response monitoring unit includes an accelerometer, strain gauge, and inclinometer;

[0018] The environmental load monitoring unit includes an anemometer, wave sensor, and current meter;

[0019] The corrosion monitoring unit includes a pH sensor and a chloride ion concentration sensor.

[0020] Preferably, the data preprocessing module uses an adaptive Kalman filter algorithm to eliminate environmental noise and a sliding window method to standardize the data.

[0021] Preferably, the feature extraction module extracts the following features simultaneously:

[0022] Time-domain features include mean, variance, kurtosis, and waveform factor.

[0023] Frequency domain characteristics: including the first five natural frequencies and their amplitudes;

[0024] Time-frequency joint features: Energy distribution features obtained through wavelet transform.

[0025] Preferably, the state assessment module adopts a bidirectional LSTM neural network model based on an attention mechanism, whose input layer receives the output of the feature extraction module and whose output layer provides the structural health index (SHI).

[0026] Preferably, the state assessment module further includes:

[0027] The online model update unit updates the model parameters periodically based on newly collected data.

[0028] The uncertainty analysis unit uses the Monte Carlo method to evaluate the confidence interval of the prediction results.

[0029] Preferably, the early warning output module implements a three-level early warning mechanism:

[0030] Level 1 Warning: A routine maintenance reminder is triggered when SHI∈(0.7,0.8);

[0031] Level 2 warning: When SHI∈(0.5,0.7), a key inspection instruction is triggered;

[0032] Level 3 warning: When SHI < 0.5, an emergency shutdown command is triggered.

[0033] A multi-dimensional health monitoring method for offshore wind turbine towers includes the following steps:

[0034] S1: Synchronously acquire structural response and environmental load data through a multi-source sensor array;

[0035] S2: Perform noise reduction and standardization preprocessing on the raw data;

[0036] S3: Extract time-domain, frequency-domain, and time-frequency joint feature parameters;

[0037] S4: Input the feature parameters into the trained health assessment model to calculate the real-time health index;

[0038] S5: Trigger the corresponding warning level based on the health index threshold.

[0039] Preferably, the training method for the health assessment model in step S4 includes:

[0040] Simulation data for different damage conditions are generated through finite element simulation.

[0041] A hybrid training set was constructed by combining actual monitoring data;

[0042] Transfer learning techniques were used to transfer the parameters of onshore wind turbine tower models to offshore models.

[0043] Preferred options also include:

[0044] Establish a digital twin model to visualize the health status of wind turbine towers in real time;

[0045] Generate a diagnostic report that includes the location of the damage and repair recommendations;

[0046] Automatically record abnormal events and generate a trend chart of health status evolution.

[0047] Compared with related technologies, the multi-dimensional parameter health monitoring system and method for offshore wind turbine towers provided by this invention have the following beneficial effects:

[0048] Comprehensive Monitoring and Data Fusion: The system integrates three-dimensional monitoring units for structural response, environmental load, and corrosion status, covering parameters such as vibration, strain, tilt, wind speed, waves, flow velocity, and pH value, achieving comprehensive perception of the multi-field coupled state of "structure-environment-materials". An adaptive Kalman filter algorithm is deployed using edge computing nodes, dynamically adjusting the filter window length according to the wave cycle to effectively separate structural response signals from environmental noise (improving the signal-to-noise ratio by more than 15dB). Multi-mode standardization methods such as Z-score, Min-Max, and Robust are used to address the issue of inconsistent dimensions in heterogeneous data, providing high-quality input for feature extraction.

[0049] Deep Feature Extraction and Intelligent Assessment: The feature extraction module simultaneously calculates time-domain (12 statistical parameters), frequency-domain (first five intrinsic frequency offsets), and time-frequency joint features (5-scale wavelet energy entropy) to form a 28-dimensional highly discriminative feature vector. The state assessment module employs a bidirectional LSTM-Attention model, automatically learning the feature weights of key time steps through an attention mechanism, solving the long-term dependency problem of traditional LSTM. The model achieves a 92.3% accuracy rate in identifying early micro-damage (such as 0.1mm-level cracks). Combined with Monte Carlo Dropout uncertainty analysis, it outputs a 95% confidence interval for the SHI value. When the interval width > 0.1, manual verification is triggered, significantly reducing the false alarm rate.

[0050] Tiered Early Warning and Rapid Response: The early warning output module implements a three-tiered dynamic early warning mechanism: Level 1 early warning (SHI 0.7-0.8) pushes daily maintenance reminders through the SCADA system, with a response time of <72 hours; Level 2 early warning (SHI 0.5-0.7) automatically dispatches drones for detailed inspections, generating a report within 24 hours; Level 3 early warning (SHI <0.5) immediately cuts off the unit's power supply and activates the emergency plan, with a response time of <5 seconds. Multi-channel early warning information push (SMS, email, audible and visual alarms) ensures that no critical information is missed, and actual tests show that emergency response efficiency is improved by 60%.

[0051] Digital Twin and Full Lifecycle Management: A digital twin model of the wind turbine tower, built using the Unity3D engine, can map SHI values ​​to the 3D model's color gradient (green→yellow→red) in real time. Damaged areas are displayed with a pulse flashing effect, and the positioning accuracy reaches 0.5 meters. The system automatically generates diagnostic reports including parameter trend charts, damage diagrams, and maintenance recommendations, supporting PDF export and online viewing. The historical data analysis module generates health status evolution curves, which, combined with finite element simulation data, achieve a remaining life prediction error of <15%, providing a quantitative basis for preventative maintenance. Attached Figure Description

[0052] Figure 1 This is a schematic diagram of the multi-dimensional parameter health monitoring system for offshore wind turbine towers provided by the present invention;

[0053] Figure 2 This is a schematic diagram of the multi-dimensional parameter health monitoring method for offshore wind turbine towers provided by the present invention. Detailed Implementation

[0054] The present invention will be further described below with reference to the accompanying drawings and embodiments.

[0055] Please refer to the following: Figures 1-2 ,in, Figure 1 This is a schematic diagram of the multi-dimensional parameter health monitoring system for offshore wind turbine towers provided by the present invention; Figure 2 This is a schematic diagram of the multi-dimensional parameter health monitoring method for offshore wind turbine towers provided by the present invention.

[0056] In the specific implementation process, such as Figures 1-2 As shown, a multi-dimensional parameter health monitoring system and method for offshore wind turbine towers is applicable to real-time assessment and early warning of the structural status of wind turbine towers in complex marine environments.

[0057] It consists of three parts: a multi-source sensor network, edge computing nodes, and a cloud-based analytics platform. The system achieves intelligent assessment and tiered early warning of structural health status by real-time acquisition, processing, and analysis of multi-dimensional parameters of the wind turbine tower structure.

[0058] Multi-source sensing module: The multi-source sensing module adopts a modular design and includes three functional units:

[0059] Structural Response Monitoring Unit: This unit is responsible for collecting dynamic response parameters of the wind turbine tower structure.

[0060] Triaxial accelerometers are installed at the top, middle and foundation sections of the tower to monitor the structural vibration characteristics. The sampling frequency of the sensors is no less than 100Hz, which can accurately capture the first five vibration modes of the structure.

[0061] Fiber optic strain gauge arrays are installed on key sections of the tower, using a distributed deployment method. A monitoring section is set up every 2 meters, and 4 strain measurement points are arranged on each section to achieve all-round monitoring of the strain field.

[0062] A high-precision inclinometer with a range of ±10° and a resolution of 0.001° is installed at the tower flange connection to detect minute structural tilt changes.

[0063] Environmental load monitoring unit:

[0064] This unit monitors the environmental loads acting on the wind turbine tower in real time: an ultrasonic anemometer is installed on the top of the nacelle to measure three-dimensional wind speed and direction, with a measurement range of 0-70 m / s and an accuracy of ±0.2 m / s.

[0065] Wave radar sensors are installed on the basic platform to monitor parameters such as wave height, wave period, and wave direction spectrum in real time.

[0066] An acoustic Doppler current profiler is deployed in the underwater section to measure the flow velocity and direction at different depths, with the monitoring range covering the entire underwater structure.

[0067] Corrosion monitoring unit:

[0068] This unit monitors the corrosion status of the structure: pH value sensors and chloride ion concentration sensors are respectively arranged in the seawater splash zone, tidal zone and full immersion zone, and the measurement is carried out using electrochemical principles, with an accuracy of ±0.1 pH and ±5% of the reading.

[0069] By installing corrosion rate probes in critical areas, based on the principle of linear polarization resistance method, the annual corrosion rate can be directly output.

[0070] Data preprocessing module:

[0071] Intelligent data preprocessing: The data preprocessing module is deployed on edge computing nodes to achieve real-time data cleaning and standardization.

[0072] Adaptive Kalman Filtering: An improved adaptive Kalman filter algorithm is used to process vibration signals. The algorithm can automatically adjust the filtering parameters according to the characteristics of environmental noise, effectively separating the structural response signal from environmental noise. For non-stationary wave load signals, a sliding window technique is used for segmented processing, with the window length adaptively adjusted according to the wave period.

[0073] Data standardization: Heterogeneous data collected from different sensors are normalized, converting all parameters to the 0-1 range. Different standardization methods are used for different data types: vibration signals are standardized using Z-score, environmental parameters are normalized using Min-Max, and corrosion data are standardized using Robust to eliminate the influence of outliers.

[0074] Feature extraction module: The feature extraction module extracts three types of feature parameters from the preprocessed data:

[0075] Time-domain characteristics: Calculate the statistical characteristics of the vibration signal, including 12 time-domain indices such as mean, variance, kurtosis, and waveform factor. These characteristics can reflect the overall vibration level and impact characteristics of the structure.

[0076] Frequency domain characteristics: The first five natural frequencies and their amplitudes of the structure are extracted using Fast Fourier Transform, and the offset rate of each frequency is calculated. Simultaneously, an autoregressive model is used to estimate the power spectrum and obtain the frequency domain energy distribution characteristics.

[0077] Time-frequency joint features: Morlet wavelet transform is used to analyze non-stationary signals, calculate wavelet energy distribution at five scales, and derive the energy entropy index for detecting local damage in structures.

[0078] Status assessment module:

[0079] Intelligent Health Assessment: The status assessment module uses a deep learning model to achieve intelligent diagnosis of structural health status.

[0080] Bidirectional LSTM-Attention Model: This model constructs a bidirectional long short-term memory neural network based on an attention mechanism. The network input is a 28-dimensional feature vector, which is processed through two layers of bidirectional LSTM, each containing 64 neurons. The attention layer automatically learns the importance weights of features at each time step, and the final output layer outputs a structural health index (SHI) in the range of 0-1 using the sigmoid function.

[0081] Online model updates: The system automatically collects new data monthly for model fine-tuning, employing transfer learning techniques to keep the underlying feature extraction parameters unchanged, adjusting only the upper-layer classifier parameters. Each update retains historical model versions, supporting model performance comparison and rollback.

[0082] Uncertainty analysis: The Monte Carlo Dropout method is used to assess the uncertainty of the prediction results. 100 forward propagation calculations are performed to output the confidence interval of the SHI value. When the interval width exceeds the threshold, manual review is triggered.

[0083] Early warning output module:

[0084] The early warning output module implements a three-level early warning mechanism:

[0085] Level 1 Warning:

[0086] When the SHI value is in the range of 0.7-0.8, the system generates a daily maintenance reminder and sends it to the maintenance personnel through the SCADA system, suggesting that a routine check be performed within 72 hours.

[0087] Level 2 warning:

[0088] When the SHI value is in the range of 0.5-0.7, the system triggers a key inspection command, automatically dispatches a drone to conduct a detailed inspection of the suspicious parts, and requires the submission of an inspection report within 24 hours.

[0089] Level 3 Warning:

[0090] When the SHI value is below 0.5, the system immediately issues an emergency shutdown command, cuts off the power supply to the unit, sends an alarm message to the operation and maintenance team, and activates the emergency plan.

[0091] Digital twin system:

[0092] 3D Visualization Platform: Developing digital twin models of wind turbine towers based on the Unity3D engine to visualize monitoring data.

[0093] Real-time status display: The 3D model is dynamically colored based on the SHI value. A healthy state is displayed in green, slight degradation in yellow, and severe damage in red. Damaged areas are shown with a pulse flashing effect for easy and rapid location.

[0094] Intelligent Report Generation: The system automatically generates diagnostic reports containing the following: trend charts of key parameter changes, damage location diagrams, maintenance recommendation lists, and remaining life predictions. Reports can be exported in PDF format and viewed online.

[0095] Historical data analysis: The system records all abnormal events, forms a health status evolution trend chart, supports querying and comparative analysis by time range, and helps to assess the pattern of structural performance degradation.

[0096] Monitoring Methodology and Procedure: The multi-dimensional parameter health monitoring method for offshore wind turbine towers includes the following steps:

[0097] Synchronous Data Acquisition: Structural response and environmental load data are synchronously acquired through a multi-source sensor array. The sampling frequency is set according to the type of monitoring parameter, with vibration signals at no less than 100Hz and environmental parameters at no less than 1Hz. All sensors are synchronized with GPS clocks, with a time synchronization accuracy better than 1 millisecond.

[0098] Data preprocessing: The raw data is denoised and standardized. Vibration signals are denoised using wavelet thresholding combined with Kalman filtering, environmental parameters are denoised using moving average filtering, and corrosion data are denoised using median filtering. All processed data are then converted to standard units.

[0099] Feature extraction: Extracting time-domain, frequency-domain, and time-frequency joint feature parameters from preprocessed data. Time-domain features include statistics such as mean and variance; frequency-domain features include natural frequencies and their amplitudes; and time-frequency features are obtained through wavelet transform to obtain energy distribution.

[0100] Health Assessment: Feature parameters are input into a trained health assessment model to calculate real-time health indices. The model is jointly trained using finite element simulation data and measured data, and transfer learning techniques are used to transfer parameters from the onshore wind turbine tower model to the offshore model.

[0101] Warning Trigger: The system triggers corresponding warning levels based on health index thresholds. It supports multiple channels for sending warning information, including SMS, email, and SCADA alarms, ensuring timely response.

[0102] The above description is merely an embodiment of the present invention and does not limit the patent scope of the present invention. Any equivalent structural or procedural transformations made based on the content of the present invention specification and drawings, or direct or indirect applications in other related technical fields, are similarly included within the patent protection scope of the present invention.

Claims

1. A multi-dimensional parameter health monitoring system for offshore wind turbine towers, characterized in that, include: Multi-source sensing module for real-time acquisition of multi-dimensional parameters of wind turbine tower structure; The data preprocessing module performs filtering and normalization on the collected data; The feature extraction module extracts time-domain and frequency-domain features from the preprocessed data; The status assessment module evaluates the health status of wind turbine towers based on machine learning models. The early warning output module generates a graded early warning signal when the evaluation result exceeds the threshold.

2. The multi-dimensional parameter health monitoring system for offshore wind turbine towers according to claim 1, characterized in that: The multi-source sensing module includes: The structural response monitoring unit includes an accelerometer, strain gauge, and inclinometer; The environmental load monitoring unit includes an anemometer, wave sensor, and current meter; The corrosion monitoring unit includes a pH sensor and a chloride ion concentration sensor.

3. The multi-dimensional parameter health monitoring system for offshore wind turbine towers according to claim 1, characterized in that: The data preprocessing module uses an adaptive Kalman filter algorithm to eliminate environmental noise and a sliding window method to standardize the data.

4. The multi-dimensional parameter health monitoring system for offshore wind turbine towers according to claim 1, characterized in that: The feature extraction module extracts the following features simultaneously: Time-domain features include mean, variance, kurtosis, and waveform factor. Frequency domain characteristics: including the first five natural frequencies and their amplitudes; Time-frequency joint features: Energy distribution features obtained through wavelet transform.

5. The multi-dimensional parameter health monitoring system for offshore wind turbine towers according to claim 1, characterized in that: The state assessment module adopts a bidirectional LSTM neural network model based on the attention mechanism. Its input layer receives the output of the feature extraction module, and its output layer gives the structural health index (SHI).

6. The multi-dimensional parameter health monitoring system for offshore wind turbine towers according to claim 5, characterized in that: The status assessment module also includes: The online model update unit updates the model parameters periodically based on newly collected data. The uncertainty analysis unit uses the Monte Carlo method to evaluate the confidence interval of the prediction results.

7. The multi-dimensional parameter health monitoring system for offshore wind turbine towers according to claim 1, characterized in that: The early warning output module implements a three-level early warning mechanism: Level 1 Warning: A routine maintenance reminder is triggered when SHI∈(0.7,0.8); Level 2 warning: When SHI∈(0.5,0.7), a key inspection instruction is triggered; Level 3 warning: When SHI < 0.5, an emergency shutdown command is triggered.

8. A method for multi-dimensional parameter health monitoring of offshore wind turbine towers, comprising the multi-dimensional parameter health monitoring system for offshore wind turbine towers as described in any one of claims 1 to 7, characterized in that: Includes the following steps: S1: Synchronously acquire structural response and environmental load data through a multi-source sensor array; S2: Perform noise reduction and standardization preprocessing on the raw data; S3: Extract time-domain, frequency-domain, and time-frequency joint feature parameters; S4: Input the feature parameters into the trained health assessment model to calculate the real-time health index; S5: Trigger the corresponding warning level based on the health index threshold.

9. The multi-dimensional parameter health monitoring system for offshore wind turbine towers according to claim 8, characterized in that: The training method for the health assessment model in step S4 includes: Simulation data for different damage conditions are generated through finite element simulation. A hybrid training set was constructed by combining actual monitoring data; Transfer learning techniques were used to transfer the parameters of onshore wind turbine tower models to offshore models.

10. The multi-dimensional parameter health monitoring system for offshore wind turbine towers according to claim 8, characterized in that: Also includes: Establish a digital twin model to visualize the health status of wind turbine towers in real time; Generate a diagnostic report that includes the location of the damage and repair recommendations; Automatically record abnormal events and generate a trend chart of health status evolution.

Images