Fan tower drum fault diagnosis method and system based on multi-domain feature fusion

Abstract

The application belongs to the technical field of fault diagnosis, and discloses a fan tower drum fault diagnosis method and system based on multi-domain feature fusion. The method comprises the following steps: constructing a tower drum bolt simulation model, obtaining a plurality of historical simulation stress data and a plurality of historical measured stress data; converting the plurality of historical simulation stress data and the corresponding plurality of historical measured stress data into a plurality of model construction samples; constructing a fan tower drum fault diagnosis model according to the plurality of model construction samples; using the tower drum bolt simulation model to obtain real-time simulation stress data and real-time measured stress data; and using the fan tower drum fault diagnosis model to perform fault diagnosis according to the real-time simulation stress data and the real-time measured stress data. The application solves the problems of insufficient accuracy, lack of effective fusion of simulation and measured data, and insufficient model capacity in the prior art.

Description

Technical Field

[0001] This invention belongs to the field of fault diagnosis technology, specifically relating to a method and system for fault diagnosis of wind turbine towers based on multi-domain feature fusion. Background Technology

[0002] With the rapid development of the wind power industry, the operational reliability and safety of wind turbine towers, as key supporting structures, are of paramount importance. Under complex loads such as bolt preload and wind loads, towers are prone to failures such as bolt loosening, bolt breakage, localized flange deformation, tower tilting, and tower cracks. These failures severely impact the stable operation and service life of wind power systems. Therefore, accurate fault diagnosis is crucial for ensuring the safe operation of towers and is of great significance for improving wind power generation efficiency and reducing maintenance costs.

[0003] Existing technologies still have many shortcomings, including:

[0004] 1) Insufficient accuracy: In existing wind turbine tower fault diagnosis technologies, some methods only analyze features of a single physical domain and fail to fully integrate multi-domain feature information such as time domain, frequency domain, and spatial domain. This results in insufficient ability to characterize complex faults, easy omission of key fault features, and inaccurate accuracy.

[0005] 2) Lack of effective integration of simulation and measured data: In existing wind turbine tower fault diagnosis technologies, fault diagnosis is often based on independent simulation or measured data. Measured data is easily affected by factors such as sensor noise, installation deviation, and environmental interference (such as stress fluctuations caused by temperature changes), resulting in unstable data quality. It is very difficult and costly to obtain measured data covering various extreme or rare fault conditions. Although simulation models can simulate stress states under ideal or specific conditions, the accuracy of the models depends on precise mechanical parameters and boundary condition settings. Actual conditions are complex and variable, and simulation results may deviate from the actual situation. Using simulation data alone for diagnosis lacks reliability. Therefore, there are relatively few technical solutions that combine simulation and measured data for diagnosis, or the integration methods are relatively simple (such as data splicing), failing to give full play to the advantages of simulation data in supplementing extreme conditions and explaining anomalies in measured data, and also failing to use measured data to calibrate and verify the potential of simulation models.

[0006] 3) Insufficient model capabilities: Existing wind turbine tower fault diagnosis technologies often use statistical algorithms or simple networks to build fault diagnosis models. When dealing with high-dimensional, strongly coupled multi-domain features, the model complexity is limited and may not be able to learn the deep complex relationships in the data. It also fails to be specifically designed for the characteristics of wind turbine tower faults. Summary of the Invention

[0007] To address the issues of insufficient accuracy, lack of effective fusion of simulation and measured data, and inadequate model capabilities in existing technologies, this invention aims to provide a wind turbine tower fault diagnosis method and system based on multi-domain feature fusion.

[0008] The technical solution adopted in this invention is as follows:

[0009] A method for fault diagnosis of wind turbine towers based on multi-domain feature fusion includes the following steps:

[0010] A simulation model of tower bolts was constructed. Under different historical wind loads and historical preloads, several historical simulated stress data were obtained, and corresponding historical measured stress data were collected.

[0011] Based on the preset multi-domain feature fusion engineering and fault judgment indicators, several historical simulation stress data and corresponding historical measured stress data are converted into several model construction samples.

[0012] Based on several model construction samples, deep learning algorithms are used to construct models and obtain a wind turbine tower fault diagnosis model based on multi-domain feature fusion.

[0013] Based on real-time wind load and real-time preload, a tower bolt simulation model is used to obtain real-time simulated stress data and collect corresponding real-time measured stress data.

[0014] Based on real-time simulated stress data and real-time measured stress data, a wind turbine tower fault diagnosis model is used to perform fault diagnosis and obtain real-time wind turbine tower fault diagnosis results.

[0015] Furthermore, a simulation model of the tower bolts was constructed. Under different historical wind loads and historical preloads, several historical simulated stress data were obtained, and corresponding historical measured stress data were collected, including the following steps:

[0016] Based on the mechanical properties of tower bolt connections, a dynamic model of tower bolts is established, and a simulation model of tower bolts is constructed using CAE tools based on the dynamic model.

[0017] Several historical SCADA wind speed data and several historical bolt preloads were input into the tower bolt simulation model, and simulations were performed under different historical wind loads and historical preloads to obtain several historical simulation stress data.

[0018] Several ultrasonic sensors were used to collect historical displacement data of tower bolts under different historical wind loads and historical preloads, and preprocessed the data to obtain several preprocessed historical displacement data.

[0019] Using Hooke's Law, several preprocessed historical displacement data are converted into corresponding historical measured stress data.

[0020] Based on several historical measured stress data, a pre-built sensor fault diagnosis and prediction model is used to generate several corresponding historical predicted stress data when a sensor fault occurs, and replace several historical measured stress data of the sensor fault.

[0021] Furthermore, based on the mechanical properties of the tower bolt connection, a dynamic model of the tower bolt is established, and based on the dynamic model, a simulation model of the tower bolt is constructed using CAE tools, including the following steps:

[0022] The tower bolt connection is simplified into a spring-mass-damping system, and corresponding wind load parameters, preload parameters, material properties and geometric parameters are set to obtain a dynamic model including dynamic equations;

[0023] Import the dynamic model, including the dynamic equations, into the CAE tool, and use the CAE tool to construct an accurate three-dimensional geometric model of the tower bolts based on the geometric parameters of the dynamic model.

[0024] Based on the material properties of the dynamic model, the material property conditions of the precise three-dimensional geometric model are defined, and the precise three-dimensional geometric model is meshed to obtain the initial tower bolt simulation model.

[0025] Based on the wind load parameters and preload parameters of the dynamic model, the wind load conditions, boundary conditions, and preload conditions of the initial tower bolt simulation model are defined to obtain the final tower bolt simulation model.

[0026] Furthermore, the pre-defined multi-domain feature fusion engineering includes time-domain feature engineering, frequency-domain feature engineering, and spatial-domain feature engineering;

[0027] Time-domain characteristic engineering includes stress mean and stress standard deviation;

[0028] Frequency domain characterization includes power spectral density and natural frequency amplitude;

[0029] Spatial domain characteristic engineering includes maximum stress and coefficient of variation;

[0030] Fault determination indicators include local bolt stress deviation, single bolt stress deviation, single bolt displacement deviation, uneven stress distribution, and comparison of natural frequency amplitude.

[0031] Wind turbine tower failure types include local flange compression deformation, loose bolts, broken bolts, tower tilting, and tower cracks.

[0032] Furthermore, based on preset multi-domain feature fusion engineering and fault determination indicators, several historical simulation stress data and corresponding historical measured stress data are converted into several model construction samples, including the following steps:

[0033] Based on the preset multi-domain feature fusion method, historical time-domain features, historical frequency-domain features, and historical spatial-domain features of historical simulated stress data and historical measured stress data are extracted, and historical multi-domain fusion features are obtained.

[0034] Based on the fault determination indicators, fault determination is performed on the historical multi-domain fusion characteristics to obtain the corresponding historical wind turbine tower fault type labels.

[0035] By adding historical wind turbine tower fault type labels to the corresponding historical multi-domain fusion features, several model construction samples are obtained.

[0036] Furthermore, the wind turbine tower fault diagnosis model is constructed based on the LSTM-AE-CNN-Attention-MLP-ISSA algorithm. The wind turbine tower fault diagnosis model includes a time-domain feature extraction module based on the LSTM algorithm, a frequency-domain feature extraction module based on the AE algorithm, a spatial-domain feature extraction module based on the CNN algorithm, a weighted fusion module based on the Attention mechanism, and a fault diagnosis module based on the MLP algorithm. The time-domain feature extraction module, the frequency-domain feature extraction module, and the spatial-domain feature extraction module are all connected to the weighted fusion module, and the weighted fusion module and the fault diagnosis module are connected in sequence.

[0037] Furthermore, based on samples from several model constructions, deep learning algorithms are used to build a model, resulting in a wind turbine tower fault diagnosis model based on multi-domain feature fusion, including the following steps:

[0038] Several model construction samples were divided into a model training sample set and a model test sample set in a 7:3 ratio;

[0039] The initial wind turbine tower fault diagnosis model was constructed using the LSTM-AE-CNN-Attention-MLP algorithm.

[0040] Input the model training sample set and optimize the initial wind turbine tower fault diagnosis model to obtain the optimized wind turbine tower fault diagnosis model.

[0041] Input the model test sample set, test the optimized wind turbine tower fault diagnosis model, and compare the output predicted wind turbine tower fault type labels with the corresponding historical wind turbine tower fault type labels to obtain the model test accuracy.

[0042] If the model's test accuracy is greater than the accuracy threshold, the final wind turbine tower fault diagnosis model will be output; otherwise, optimization training will continue.

[0043] Furthermore, based on the real-time wind load and real-time preload, a tower bolt simulation model is used to obtain real-time simulated stress data, and corresponding real-time measured stress data is collected, including the following steps:

[0044] Real-time SCADA wind speed data and real-time bolt preload are input into the tower bolt simulation model, and simulation is performed under real-time wind load and real-time preload to obtain real-time simulation stress data.

[0045] Ultrasonic sensors are used to collect real-time displacement data of tower bolts under real-time wind load and real-time preload, and the data is preprocessed to obtain preprocessed real-time displacement data.

[0046] Using Hooke's Law, the preprocessed real-time displacement data is converted into corresponding real-time measured stress data;

[0047] Based on real-time measured stress data, a pre-built sensor fault diagnosis and prediction model is used to generate corresponding real-time predicted stress data when a sensor fault occurs, and replace the real-time measured stress data of the sensor fault.

[0048] Furthermore, based on real-time simulated stress data and real-time measured stress data, a wind turbine tower fault diagnosis model is used to perform fault diagnosis and obtain real-time wind turbine tower fault diagnosis results, including the following steps:

[0049] The real-time simulated stress data and the real-time measured stress data are synchronized in time to obtain synchronized real-time simulated stress data and synchronized real-time measured stress data. The synchronized real-time simulated stress data and synchronized real-time measured stress data are then input into the wind turbine tower fault diagnosis model.

[0050] Using the time-domain feature extraction module, frequency-domain feature extraction module and spatial-domain feature extraction module of the wind turbine tower fault diagnosis model, real-time time-domain features, real-time frequency-domain features and real-time spatial-domain features of the real-time simulated stress data and the real-time measured stress data after synchronization are extracted.

[0051] Based on dynamic attention weights, the weighted fusion module of the wind turbine tower fault diagnosis model is used to perform weighted fusion of real-time time domain features, real-time frequency domain features, and real-time spatial domain features to obtain real-time multi-domain fusion features.

[0052] Based on the real-time multi-domain fusion characteristics, the fault diagnosis module of the wind turbine tower fault diagnosis model is used to perform fault diagnosis and obtain real-time wind turbine tower fault diagnosis results.

[0053] A wind turbine tower fault diagnosis system based on multi-domain feature fusion is used to implement a wind turbine tower fault diagnosis method. The system includes a simulation model construction unit, a sample generation unit, a fault diagnosis model construction unit, a simulation data generation unit, and a fault diagnosis execution unit connected in sequence.

[0054] The beneficial effects of this invention are as follows:

[0055] This invention provides a wind turbine tower fault diagnosis method and system based on multi-domain feature fusion. It integrates feature information from the time, frequency, and spatial domains, enabling a more comprehensive and in-depth characterization of the complex states of wind turbine tower bolt connections. This effectively avoids the limitations of single-physical-domain analysis, minimizes the omission of key fault features, and significantly improves the accuracy and reliability of fault diagnosis, meeting higher engineering requirements. By combining simulated and measured stress data of tower bolts, the simulation model can generate stress data covering various extreme or rare fault conditions at low cost, effectively supplementing the deficiencies of measured data and explaining abnormal fluctuations in measurements. Furthermore, the measured data is not only used for model training but also for implicit calibration of the simulation model. This method improves the real-world relevance of simulation results. Instead of simply piecing together data, it uses both simulation and experimental data as the foundation for constructing samples. This allows the model to learn more robust features through the synergistic effect of simulation and experimental data, overcoming the limitations of using either data source alone (unstable quality of experimental data and potential deviations between simulation and actual results), thus enhancing the reliability of diagnostic results. Furthermore, it employs a combination of advanced deep learning algorithms to construct a complex model specifically designed for the characteristics of wind turbine tower faults. This model effectively handles high-dimensional, strongly coupled multi-domain feature data, learning deep-seated complex nonlinear relationships within the data. Compared to traditional statistical algorithms or simple networks, the fault diagnosis model possesses stronger representation and generalization capabilities, enabling it to capture fault characteristics more accurately.

[0056] Other beneficial effects of the present invention will be further explained in the specific embodiments. Attached Figure Description

[0057] Figure 1 This is a flowchart of the wind turbine tower fault diagnosis method based on multi-domain feature fusion in this invention.

[0058] Figure 2 This is a structural block diagram of the wind turbine tower fault diagnosis system based on multi-domain feature fusion in this invention. Detailed Implementation

[0059] The present invention will be further explained below with reference to the accompanying drawings and specific embodiments.

[0060] Example 1:

[0061] like Figure 1As shown in the figure, this embodiment provides a wind turbine tower fault diagnosis method based on multi-domain feature fusion, including the following steps:

[0062] S1: Construct a simulation model of the tower bolts, obtain several historical simulated stress data under different historical wind loads and historical preloads, and collect corresponding historical measured stress data, including the following steps:

[0063] S1-1: Based on the mechanical properties of the tower bolt connection, establish a dynamic model of the tower bolts, and based on the dynamic model, use Computer-Aided Engineering (CAE) tools to construct a simulation model of the tower bolts, including the following steps:

[0064] S1-1-1: The tower bolt connection is simplified into a spring-mass-damping system, and the corresponding wind load parameters, preload parameters, material properties and geometric parameters are set to obtain a dynamic model including dynamic equations;

[0065] S1-1-2: Import the dynamic model, including the dynamic equations, into the CAE tool, and use the CAE tool to construct an accurate three-dimensional geometric model of the tower bolts based on the geometric parameters of the dynamic model.

[0066] S1-1-3: Based on the material properties of the dynamic model, define the material property conditions of the precise three-dimensional geometric model, and perform mesh generation on the precise three-dimensional geometric model to obtain the initial tower bolt simulation model;

[0067] S1-1-4: Based on the wind load parameters and preload parameters of the dynamic model, define the wind load conditions, boundary conditions, and preload conditions of the initial tower bolt simulation model to obtain the final tower bolt simulation model.

[0068] S1-2: Input several historical wind speed data from the Supervisory Control and Data Acquisition (SCADA) system and several historical bolt preloads into the tower bolt simulation model, and simulate under different historical wind loads and historical preloads to obtain several historical simulation stress data.

[0069] S1-3: Using several ultrasonic sensors, collect several historical displacement data of tower bolts under different historical wind loads and historical preloads, and preprocess them to obtain several preprocessed historical displacement data.

[0070] S1-4: Using Hooke's Law, several preprocessed historical displacement data are converted into corresponding historical measured stress data.

[0071] The formula is:

[0072] .

[0073] In the formula, For time Measured stress data; The elastic modulus of the material; This is the original length of the bolt; For time Displacement data; For time indication;

[0074] S1-5: Based on several historical measured stress data, using a pre-built sensor fault diagnosis and prediction model, when a sensor fault occurs, generate several corresponding historical predicted stress data and replace several historical measured stress data of the sensor fault.

[0075] The sensor fault diagnosis and prediction model is built based on the Support Vector Machine (SVM)-Transformer algorithm, and the sensor fault diagnosis and prediction model includes a sensor fault diagnosis module built based on the SVM algorithm and a sensor data prediction module built based on the Transformer algorithm, which are connected in sequence.

[0076] S1-5: Based on several historical measured stress data, using a pre-built sensor fault diagnosis and prediction model, when a sensor fault occurs, generate several corresponding historical predicted stress data and replace several historical measured stress data of the sensor fault, including the following steps:

[0077] S1-5-1: Input the historical measured stress data into the sensor fault diagnosis module of the sensor fault diagnosis and prediction model for judgment. If the sensor fault is determined to have occurred, proceed to the next step; otherwise, traverse the next historical measured stress data.

[0078] S1-5-2: The historical measured stress data sequence, which includes the historical measured stress data corresponding to the fault history, and the corresponding time-coded sequence containing periodic features are spliced ​​together to form a historical input matrix that integrates time and stress information;

[0079] S1-5-3: Sensor data prediction module using sensor fault diagnosis and prediction model, extracts historical high-order features of stress and time coupling of historical input matrix;

[0080] S1-5-4: Based on historical high-order characteristics, perform sensor data prediction to obtain historical predicted stress data for the next 24 hours;

[0081] S1-5-5: Traverse all historical measured stress data of all faults, corresponding historical predicted stress data, and replace several historical measured stress data of sensor faults.

[0082] S2: Based on the preset multi-domain feature fusion engineering and fault judgment indicators, convert several historical simulation stress data and corresponding historical measured stress data into several model construction samples;

[0083] The pre-defined multi-domain feature fusion engineering includes time-domain feature engineering, frequency-domain feature engineering, and spatial-domain feature engineering.

[0084] Time-domain characteristic engineering includes stress mean and stress standard deviation;

[0085] The formula is:

[0086]

[0087] In the formula, This represents the average stress value. Total number of hours; For time Stress data; For time indication;

[0088]

[0089] In the formula, The standard deviation of stress; This represents the total number of stress data. This is a stress data indicator; For the first Stress data; This represents the average stress value.

[0090] Frequency domain characterization includes power spectral density and natural frequency amplitude;

[0091] The formula is:

[0092]

[0093] In the formula, Power spectral density; Total number of hours; For time Stress data; f is the time indicator; j is the frequency indicator; e is the complex parameter; and e is a natural number.

[0094]

[0095] In the formula, The amplitudes of simulated stress data and measured stress data at the natural frequency; This is the amplitude calculation function; The first-order natural frequency of the tower;

[0096] Spatial domain characteristic engineering includes maximum stress and coefficient of variation;

[0097] The formula is:

[0098]

[0099] In the formula, This represents the maximum stress value from the simulated stress data; This represents the maximum stress value from the measured stress data; For the first Simulation stress data and the first Measured stress data;

[0100]

[0101] In the formula, The stress variation coefficient is the measured stress data. The standard deviation of the measured stress data; This represents the average stress value from the measured stress data.

[0102] Fault determination indicators include local bolt stress deviation, single bolt stress deviation, single bolt displacement deviation, uneven stress distribution, and comparison of natural frequency amplitude.

[0103] Wind turbine tower failure types include local flange compression deformation, loose bolts, broken bolts, tower tilting, and tower cracks;

[0104] Local bolt stress deviation: Calculate the measured stress data of a single bolt in a localized area. Stress mean compared with simulated stress data The deviation, where k is the measured stress data indication:

[0105]

[0106] If the conditions are met, it is determined to be localized flange compression deformation;

[0107] Statistical analysis of a large amount of simulation and actual measurement data revealed that under normal operating conditions, the fluctuation range of bolt stress is relatively small, and the deviation between the measured stress and the average value of the simulation group is usually within a small range. However, when faults such as local flange compression occur, this deviation increases significantly. Through analysis and comparison of data under different fault conditions, 15% was determined as a threshold that can effectively distinguish between normal and fault conditions.

[0108] Stress deviation of a single bolt: By comparing the measured stress data and simulated stress data of a single bolt, among which, These are measured stress data. For simulation stress data:

[0109]

[0110] If the conditions are met, the bolt is determined to be loose.

[0111] Single bolt displacement deviation: Compare with the measured displacement data of a single bolt. With simulated displacement data :

[0112]

[0113] If the condition is met, it is determined that the bolt is broken. The bolt displacement data reflects the change in bolt length. The sensor accuracy is 0.001. Under normal circumstances, the error will fluctuate, but will not exceed 0.01.

[0114] Uneven stress distribution: By calculating the coefficient of variation of the measured stress:

[0115] >0.1

[0116] If the conditions are met, then the tower is determined to be tilted, and the stress variation coefficient is... It is a dimensionless index calculated by dividing the standard deviation by the mean, used to quantify the uniformity of stress distribution in a bolt group; the threshold is 0.1. This is based on statistical analysis of a large amount of measured data: during normal operation... Less than 0.8, 0.08 when slightly tilted. <0.1, significant tilt >0.1;

[0117] Cross-validation was performed using tilt sensor data: Tilt angle verification: If the tower tilt angle... Where H is the tower height, it supports tower tilting faults; the deviation of the wind turbine tower tilt rate is no more than 5% of the wind turbine tower height;

[0118] Natural frequency amplitude comparison: The amplitude of the measured stress data at the crack characteristic frequency of 10-20kHz. The amplitude is compared with three times the background noise amplitude as a crack detection threshold. If the threshold is met, the tower is identified as having a crack. When a tower crack forms, it changes the stiffness and damping characteristics of the structure, leading to local stress concentration. When the structure vibrates, the crack surface will generate characteristic vibration components in a specific high-frequency band of 10-20kHz due to relative slippage, friction, or stress wave reflection and scattering. These high-frequency signals are usually related to the geometric size and dynamic response characteristics of the crack, which are different from the low-frequency modes of the overall tower vibration.

[0119] Cross-validation was performed using accelerometer data: if the root mean square of the acceleration... Exceeding the threshold This supports tower crack failure, where statistical analysis (such as calculating the mean) is used. and standard deviation ), threshold Set as , This is the threshold coefficient;

[0120] The cross-validation mechanism uses multi-source sensor data to cross-check each other, avoiding the influence of single data deviations, ensuring more reliable fault diagnosis results, effectively reducing false and false diagnoses, and providing more accurate fault diagnosis support for the safe operation of wind turbine towers.

[0121] Based on preset multi-domain feature fusion engineering and fault judgment indicators, several historical simulation stress data and corresponding historical measured stress data are converted into several model construction samples, including the following steps:

[0122] S2-1: Based on the preset multi-domain feature fusion process, extract the historical time domain features, historical frequency domain features, and historical spatial domain features of historical simulated stress data and historical measured stress data, and obtain the historical multi-domain fusion features;

[0123] S2-2: Based on the fault determination indicators, perform fault determination on the historical multi-domain fusion features to obtain the corresponding historical wind turbine tower fault type labels;

[0124] S2-3: Add the historical wind turbine tower fault type labels to the corresponding historical multi-domain fusion features to obtain several model construction samples;

[0125] S3: Based on several model construction samples, use deep learning algorithms to construct models and obtain a wind turbine tower fault diagnosis model based on multi-domain feature fusion;

[0126] The wind turbine tower fault diagnosis model is constructed based on the Long Short-Term Memory (LSTM) network-Autoencoder (AE)-Convolutional Neural Network (CNN)-Attention-Multilayer Perceptron (MLP) algorithm. The wind turbine tower fault diagnosis model includes a time-domain feature extraction module based on the LSTM algorithm, a frequency-domain feature extraction module based on the AE algorithm, a spatial-domain feature extraction module based on the CNN algorithm, a weighted fusion module based on the Attention mechanism, and a fault diagnosis module based on the MLP algorithm. The time-domain feature extraction module, the frequency-domain feature extraction module, and the spatial-domain feature extraction module are all connected to the weighted fusion module, and the weighted fusion module and the fault diagnosis module are connected in sequence.

[0127] Based on several model construction samples, deep learning algorithms are used to construct models, resulting in a wind turbine tower fault diagnosis model based on multi-domain feature fusion, including the following steps:

[0128] S3-1: Divide several model construction samples into a model training sample set and a model test sample set in a 7:3 ratio;

[0129] S3-2: Use the LSTM-AE-CNN-Attention-MLP algorithm to build an initial wind turbine tower fault diagnosis model;

[0130] S3-3: Input the model training sample set, optimize and train the initial wind turbine tower fault diagnosis model to obtain the optimized wind turbine tower fault diagnosis model;

[0131] S3-4: Input the model test sample set, test the optimized wind turbine tower fault diagnosis model, and compare the output predicted wind turbine tower fault type labels with the corresponding historical wind turbine tower fault type labels to obtain the model test accuracy.

[0132] S3-5: If the model test accuracy is greater than the accuracy threshold, output the final wind turbine tower fault diagnosis model; otherwise, continue optimization training.

[0133] S4: Based on the real-time wind load and real-time preload, use the tower bolt simulation model to obtain real-time simulated stress data and collect the corresponding real-time measured stress data, including the following steps:

[0134] S4-1: Input real-time SCADA wind speed data and real-time bolt preload into the tower bolt simulation model, and simulate under real-time wind load and real-time preload to obtain real-time simulation stress data;

[0135] S4-2: Using ultrasonic sensors, real-time displacement data of tower bolts under real-time wind load and real-time preload is collected and preprocessed to obtain preprocessed real-time displacement data.

[0136] S4-3: Using Hooke's Law, the preprocessed real-time displacement data is converted into corresponding real-time measured stress data;

[0137] S4-4: Based on real-time measured stress data, using a pre-built sensor fault diagnosis and prediction model, when a sensor fault occurs, generate corresponding real-time predicted stress data and replace the real-time measured stress data of the sensor fault.

[0138] S5: Based on real-time simulated stress data and real-time measured stress data, use the wind turbine tower fault diagnosis model to perform fault diagnosis and obtain real-time wind turbine tower fault diagnosis results, including the following steps:

[0139] S5-1: Synchronize the real-time simulated stress data and the real-time measured stress data to obtain synchronized real-time simulated stress data and synchronized real-time measured stress data, and input the synchronized real-time simulated stress data and synchronized real-time measured stress data into the wind turbine tower fault diagnosis model.

[0140] S5-2: Using the time-domain feature extraction module, frequency-domain feature extraction module and spatial-domain feature extraction module of the wind turbine tower fault diagnosis model, extract the real-time time-domain features, real-time frequency-domain features and real-time spatial-domain features of the real-time simulated stress data and the real-time measured stress data after synchronization.

[0141] S5-3: Based on the dynamic attention weight, the weighted fusion module of the wind turbine tower fault diagnosis model is used to perform weighted fusion of real-time time domain features, real-time frequency domain features and real-time spatial domain features to obtain real-time multi-domain fusion features.

[0142] S5-4: Based on the real-time multi-domain fusion characteristics, the fault diagnosis module of the wind turbine tower fault diagnosis model is used to perform fault diagnosis and obtain real-time wind turbine tower fault diagnosis results.

[0143] Example 2:

[0144] like Figure 2 As shown, this embodiment provides a wind turbine tower fault diagnosis system based on multi-domain feature fusion, which is used to implement a wind turbine tower fault diagnosis method. The system includes a simulation model construction unit, a sample generation unit, a fault diagnosis model construction unit, a simulation data generation unit, and a fault diagnosis execution unit connected in sequence.

[0145] The simulation model building unit is used to build a simulation model of tower bolts, obtain several historical simulation stress data under different historical wind loads and historical preload, and collect several corresponding historical measured stress data.

[0146] The sample generation unit is used to convert several historical simulation stress data and corresponding historical measured stress data into several model construction samples based on preset multi-domain feature fusion engineering and fault judgment indicators.

[0147] The fault diagnosis model building unit is used to build samples based on several models, use deep learning algorithms to build models, and obtain a wind turbine tower fault diagnosis model based on multi-domain feature fusion.

[0148] The simulation data generation unit is used to obtain real-time simulated stress data and collect corresponding real-time measured stress data based on the real-time wind load and real-time preload using the tower bolt simulation model.

[0149] The fault diagnosis execution unit is used to perform fault diagnosis based on real-time simulated stress data and real-time measured stress data using a wind turbine tower fault diagnosis model, and obtain real-time wind turbine tower fault diagnosis results.

[0150] This invention provides a wind turbine tower fault diagnosis method and system based on multi-domain feature fusion. It integrates feature information from the time, frequency, and spatial domains, enabling a more comprehensive and in-depth characterization of the complex states of wind turbine tower bolt connections. This effectively avoids the limitations of single-physical-domain analysis, minimizes the omission of key fault features, and significantly improves the accuracy and reliability of fault diagnosis, meeting higher engineering requirements. By combining simulated and measured stress data of tower bolts, the simulation model can generate stress data covering various extreme or rare fault conditions at low cost, effectively supplementing the deficiencies of measured data and explaining abnormal fluctuations in measurements. Furthermore, the measured data is not only used for model training but also for implicit calibration of the simulation model. This method improves the real-world relevance of simulation results. Instead of simply piecing together data, it uses both simulation and experimental data as the foundation for constructing samples. This allows the model to learn more robust features through the synergistic effect of simulation and experimental data, overcoming the limitations of using either data source alone (unstable quality of experimental data and potential deviations between simulation and actual results), thus enhancing the reliability of diagnostic results. Furthermore, it employs a combination of advanced deep learning algorithms to construct a complex model specifically designed for the characteristics of wind turbine tower faults. This model effectively handles high-dimensional, strongly coupled multi-domain feature data, learning deep-seated complex nonlinear relationships within the data. Compared to traditional statistical algorithms or simple networks, the fault diagnosis model possesses stronger representation and generalization capabilities, enabling it to capture fault characteristics more accurately.

[0151] This invention is not limited to the optional embodiments described above, and anyone can derive other various forms of products based on the inspiration of this invention. The specific embodiments described above should not be construed as limiting the scope of protection of this invention; the scope of protection of this invention should be determined by the claims, and the specification can be used to interpret the claims.

Claims

1. A method for fault diagnosis of wind turbine towers based on multi-domain feature fusion, characterized in that: Includes the following steps: S1; Construct a simulation model of the tower bolts, obtain several historical simulated stress data under different historical wind loads and historical preloads, and collect several corresponding historical measured stress data, including the following steps: Based on the mechanical properties of tower bolt connections, a dynamic model of tower bolts is established, and a simulation model of tower bolts is constructed using CAE tools based on the dynamic model. Several historical SCADA wind speed data and several historical bolt preloads were input into the tower bolt simulation model, and simulations were performed under different historical wind loads and historical preloads to obtain several historical simulation stress data. Several ultrasonic sensors were used to collect historical displacement data of tower bolts under different historical wind loads and historical preloads, and preprocessed the data to obtain several preprocessed historical displacement data. Using Hooke's Law, several preprocessed historical displacement data are converted into corresponding historical measured stress data. Based on several historical measured stress data, a pre-built sensor fault diagnosis and prediction model is used to generate several corresponding historical predicted stress data when a sensor fault occurs, and replace several historical measured stress data of the sensor fault. S2; Based on the preset multi-domain feature fusion engineering and fault judgment indicators, convert several historical simulation stress data and corresponding several historical measured stress data into several model construction samples; The preset multi-domain feature fusion engineering includes time-domain feature engineering, frequency-domain feature engineering, and spatial-domain feature engineering; The time-domain characteristic engineering includes the mean stress and the standard deviation of stress; The frequency domain characteristic engineering includes power spectral density and natural frequency amplitude; The spatial domain characteristic engineering includes the maximum stress value and the coefficient of variation; The fault determination indicators include local bolt stress deviation, single bolt stress deviation, single bolt displacement deviation, uneven stress distribution, and comparison of natural frequency amplitude. Wind turbine tower failure types include local flange compression deformation, loose bolts, broken bolts, tower tilting, and tower cracks; S3; Based on several model construction samples, deep learning algorithms are used to construct models and obtain a wind turbine tower fault diagnosis model based on multi-domain feature fusion; The wind turbine tower fault diagnosis model is constructed based on the LSTM-AE-CNN-Attention-MLP algorithm. The wind turbine tower fault diagnosis model includes a time-domain feature extraction module based on the LSTM algorithm, a frequency-domain feature extraction module based on the AE algorithm, a spatial-domain feature extraction module based on the CNN algorithm, a weighted fusion module based on the Attention mechanism, and a fault diagnosis module based on the MLP algorithm. The time-domain feature extraction module, the frequency-domain feature extraction module, and the spatial-domain feature extraction module are all connected to the weighted fusion module. The weighted fusion module and the fault diagnosis module are connected in sequence. S4; Based on the real-time wind load and real-time preload, use the tower bolt simulation model to obtain real-time simulated stress data and collect the corresponding real-time measured stress data; S5； Based on real-time simulated stress data and real-time measured stress data, a wind turbine tower fault diagnosis model is used to perform fault diagnosis and obtain real-time wind turbine tower fault diagnosis results.

2. The wind turbine tower fault diagnosis method based on multi-domain feature fusion according to claim 1, characterized in that: Based on the mechanical properties of tower bolt connections, a dynamic model of the tower bolts is established. Then, using CAE tools, a simulation model of the tower bolts is constructed, including the following steps: The tower bolt connection is simplified into a spring-mass-damping system, and corresponding wind load parameters, preload parameters, material properties and geometric parameters are set to obtain a dynamic model including dynamic equations; Import the dynamic model, including the dynamic equations, into the CAE tool, and use the CAE tool to construct an accurate three-dimensional geometric model of the tower bolts based on the geometric parameters of the dynamic model. Based on the material properties of the dynamic model, the material property conditions of the precise three-dimensional geometric model are defined, and the precise three-dimensional geometric model is meshed to obtain the initial simulation model of the tower bolts. Based on the wind load parameters and preload parameters of the dynamic model, the wind load conditions, boundary conditions, and preload conditions of the initial tower bolt simulation model are defined to obtain the final tower bolt simulation model.

3. The wind turbine tower fault diagnosis method based on multi-domain feature fusion according to claim 2, characterized in that: Based on preset multi-domain feature fusion engineering and fault judgment indicators, several historical simulation stress data and corresponding historical measured stress data are converted into several model construction samples, including the following steps: Based on the preset multi-domain feature fusion algorithm, historical time-domain features, historical frequency-domain features, and historical spatial-domain features of historical simulated stress data and historical measured stress data are extracted, and historical multi-domain fusion features are obtained. Based on the fault determination indicators, fault determination is performed on the historical multi-domain fusion characteristics to obtain the corresponding historical wind turbine tower fault type labels. By adding historical wind turbine tower fault type labels to the corresponding historical multi-domain fusion features, several model construction samples are obtained.

4. The wind turbine tower fault diagnosis method based on multi-domain feature fusion according to claim 3, characterized in that: Based on several model construction samples, deep learning algorithms are used to construct models, resulting in a wind turbine tower fault diagnosis model based on multi-domain feature fusion, including the following steps: Several model construction samples were divided into a model training sample set and a model test sample set in a 7:3 ratio; The initial wind turbine tower fault diagnosis model was constructed using the LSTM-AE-CNN-Attention-MLP algorithm. Input the model training sample set and optimize the initial wind turbine tower fault diagnosis model to obtain the optimized wind turbine tower fault diagnosis model. Input the model test sample set, test the optimized wind turbine tower fault diagnosis model, and compare the output predicted wind turbine tower fault type labels with the corresponding historical wind turbine tower fault type labels to obtain the model test accuracy. If the model's test accuracy is greater than the accuracy threshold, the final wind turbine tower fault diagnosis model will be output; otherwise, optimization training will continue.

5. The wind turbine tower fault diagnosis method based on multi-domain feature fusion according to claim 4, characterized in that: Based on real-time wind load and real-time preload, a tower bolt simulation model is used to obtain real-time simulated stress data, and corresponding real-time measured stress data is collected, including the following steps: Real-time SCADA wind speed data and real-time bolt preload are input into the tower bolt simulation model, and simulation is performed under real-time wind load and real-time preload to obtain real-time simulation stress data. Ultrasonic sensors are used to collect real-time displacement data of tower bolts under real-time wind load and real-time preload, and the data is preprocessed to obtain preprocessed real-time displacement data. Using Hooke's Law, the preprocessed real-time displacement data is converted into corresponding real-time measured stress data; Based on real-time measured stress data, a pre-built sensor fault diagnosis and prediction model is used to generate corresponding real-time predicted stress data when a sensor fault occurs, and replace the real-time measured stress data of the sensor fault.

6. The wind turbine tower fault diagnosis method based on multi-domain feature fusion according to claim 5, characterized in that: Based on real-time simulated stress data and real-time measured stress data, a wind turbine tower fault diagnosis model is used to perform fault diagnosis and obtain real-time wind turbine tower fault diagnosis results, including the following steps: The real-time simulated stress data and the real-time measured stress data are synchronized in time to obtain synchronized real-time simulated stress data and synchronized real-time measured stress data. The synchronized real-time simulated stress data and synchronized real-time measured stress data are then input into the wind turbine tower fault diagnosis model. Using the time-domain feature extraction module, frequency-domain feature extraction module and spatial-domain feature extraction module of the wind turbine tower fault diagnosis model, real-time time-domain features, real-time frequency-domain features and real-time spatial-domain features of the real-time simulated stress data and the real-time measured stress data after synchronization are extracted. Based on dynamic attention weights, the weighted fusion module of the wind turbine tower fault diagnosis model is used to perform weighted fusion of real-time time domain features, real-time frequency domain features, and real-time spatial domain features to obtain real-time multi-domain fusion features. Based on the real-time multi-domain fusion characteristics, the fault diagnosis module of the wind turbine tower fault diagnosis model is used to perform fault diagnosis and obtain real-time wind turbine tower fault diagnosis results.

7. A wind turbine tower fault diagnosis system based on multi-domain feature fusion, used to implement the wind turbine tower fault diagnosis method as described in any one of claims 1-6, characterized in that: The system comprises a simulation model building unit, a sample generation unit, a fault diagnosis model building unit, a simulation data generation unit, and a fault diagnosis execution unit, which are connected in sequence.

Images