Acoustic characteristic analysis-based wind driven generator blade crack detection system

Abstract

The invention relates to the technical field of wind driven generator detection, in particular to a wind driven generator blade crack detection system based on acoustic characteristic analysis, which comprises an acoustic signal acquisition module, a material fatigue state monitoring module, a signal separation and identification module and a crack positioning module, wherein the acoustic signal acquisition module acquires acoustic signals generated in the blade operation process; the material fatigue state monitoring module is used for comparing the material characteristic change with a material damage database and judging the material fatigue state of the current blade; the signal separation and identification module is used for distinguishing the acoustic signal in normal operation from the abnormal acoustic signal; distinguishing the crack signal from the material fatigue signal; and the crack positioning module determines the position of the crack through a triangulation positioning algorithm. According to the invention, through collaborative analysis of the material behavior and the acoustic signal, the separation precision of the abnormal signal is greatly improved, so that the crack signal is detected more accurately.

Description

Technical Field

[0001] The invention relates to the technical field of wind turbine detection, and in particular to a wind turbine blade crack detection system based on acoustic feature analysis. Background Art

[0002] With the rapid development of wind power generation technology, wind turbine blades, as the core components of wind power systems, are crucial to the reliable operation of the entire wind power system in terms of operational safety and stability. However, during long-term operation, wind turbine blades are prone to cracks or other forms of damage due to the complex wind loads, gravity and material factors. If these cracks cannot be detected and handled in a timely manner, they may cause serious structural failure, resulting in reduced power generation efficiency or even equipment scrapping. Therefore, how to efficiently and accurately detect cracks in wind turbine blades has become a technical problem that needs to be urgently solved in the industry.

[0003] At present, blade crack detection technology is mainly divided into two categories: visual detection-based methods and acoustic detection-based methods. Visual detection technology usually uses cameras or laser scanners to monitor the blade surface to detect surface cracks. However, visual detection is often limited by ambient light, viewing angle and blade surface conditions, making it difficult to effectively detect internal cracks. In addition, wind turbine blades are usually large, and visual detection is costly and has poor real-time performance. In contrast, crack detection technology based on acoustic features has become one of the mainstream methods in current research because it can perform non-destructive detection of internal cracks in blades and has significant advantages in environmental adaptability and real-time performance. Existing acoustic crack detection technology mainly collects sound wave signals during blade operation through acoustic sensors, and uses spectrum analysis or filtering to analyze abnormal signals to detect the presence of cracks. However, the existing technology still has many technical bottlenecks in terms of signal separation accuracy and the ability to distinguish crack signals from material fatigue signals. Specifically, existing systems often rely on traditional linear filtering algorithms, which are difficult to effectively separate complex abnormal acoustic signals, especially when facing tiny cracks and complex material fatigue states, and the detection accuracy is insufficient. Summary of the invention

[0004] The invention provides a wind turbine blade crack detection system based on acoustic feature analysis.

[0005] The wind turbine blade crack detection system based on acoustic feature analysis includes an acoustic signal acquisition module, a material fatigue state monitoring module, a signal separation and identification module, and a crack location module, wherein:

[0006] The acoustic signal acquisition module collects the acoustic signals generated during the operation of the wind turbine blades in real time by installing acoustic sensors at multiple positions of the wind turbine blades;

[0007] The material fatigue state monitoring module uses a multimodal sensor to monitor the material property changes of the blade material in real time. The material property changes include stress changes and temperature gradients. The material fatigue state monitoring module has a built-in material damage database, compares the material property changes with the material damage database, and combines the wind turbine blade operation history data to determine the material fatigue state of the current blade;

[0008] The signal separation and recognition module performs multi-level analysis on the collected acoustic signals through a nonlinear signal decomposition algorithm and a self-supervised learning model, and decomposes the acoustic signals using the Hilbert-Huang Transform (HHT) to distinguish between normal operating acoustic signals and abnormal acoustic signals; the abnormal acoustic signals also include crack signals and material fatigue signals, and based on the material fatigue state data output by the fatigue state monitoring module, the double matching method of the acoustic feature time-frequency diagram and the strain change diagram is used to distinguish the crack signal from the material fatigue signal;

[0009] After distinguishing the crack signal, the crack location module combines the data of the multi-point acoustic sensor and determines the crack position through a triangulation positioning algorithm.

[0010] Optionally, the acoustic signal acquisition module installs array acoustic sensors at multiple predetermined positions on the wind turbine blade, wherein the predetermined positions include the root, middle section and tip of the blade, to ensure full coverage of key structural areas of the blade. Each acoustic sensor is interconnected through a wireless communication network to form a distributed acoustic signal acquisition unit, which collects in real time the acoustic signals generated by the blade under different wind speeds, wind directions and load conditions.

[0011] Optionally, the material fatigue state monitoring module collects data on stress changes and temperature gradient changes of the blades in real time through a multimodal sensor array installed on the surface and inside of the blades. The multimodal sensor array includes strain sensors and thermistors, which capture the fatigue state of the blade material in multiple dimensions. The material damage database built into the material fatigue state monitoring module stores damage thresholds based on a variety of blade material types and under different operating environment conditions. By matching the real-time collected material property changes with the damage thresholds in the material damage database, and combining the operating history data of the wind turbine blades, a multivariate regression analysis method is used to evaluate the current material fatigue state of the blades in real time.

[0012] Optionally, the multivariate regression analysis method is expressed as:

[0013] Y=β 0 +β 1 σ+β 2 T g +∈, where Y is the fatigue state value, indicating the degree to which the material is close to fatigue damage, σ is the stress value calculated by the stress-strain relationship, Tg is the temperature gradient, which indicates the local temperature change of the blade, collected by the thermal sensor, β 0 is a constant term, β 1 , β 2 is the regression coefficient obtained through historical data training, and ∈ is the error term.

[0014] Optionally, the stress-strain relationship is calculated as: σ=E·ε, where E is the Young's modulus of the material, ε is the strain value measured by the strain sensor, and the fatigue damage state is evaluated by calculating the fatigue state value Y through a multivariate regression model, which is then compared with the material damage threshold T damage For comparison: D fatigue is the evaluation value of the material fatigue damage state, when D fatigue ≥1, it means that the material has reached the critical point of fatigue. fatigue When <1, the material is in a safe state.

[0015] Optionally, the signal separation and identification module uses empirical mode decomposition and Hilbert spectrum analysis in the Hilbert-Huang transform to decompose the acoustic signal in the time domain and frequency domain, extract each intrinsic mode function (IMF), identify the tiny frequency and energy change characteristics in the acoustic signal, analyze the frequency components and energy distribution of the decomposed signal, and then use a classifier to automatically distinguish between normally operating acoustic signals and abnormal acoustic signals.

[0016] Optionally, the classifier is trained in conjunction with a self-supervised learning model, and the signal features (frequency and energy changes) obtained by Hilbert-Huang transform decomposition are input into the self-supervised learning model. The self-supervised learning model is trained through contrast learning tasks. The model automatically learns how to distinguish between normal and abnormal signals without manually labeled data. Through training, a feature representation of the signal is generated, which is then used to automatically identify new signals and determine whether they are abnormal signals.

[0017] Optionally, the use of a dual matching method of acoustic feature time-frequency graph and strain change graph to distinguish crack signals from material fatigue signals specifically includes:

[0018] Acoustic feature time-frequency diagram generation: In the signal separation and identification module, after the Hilbert-Huang transform, the acoustic signal is decomposed into multiple intrinsic mode functions, and the instantaneous frequency and amplitude of the acoustic signal are extracted through Hilbert spectrum analysis. The acoustic feature time-frequency diagram is generated based on the extracted instantaneous frequency and amplitude of the acoustic signal, which shows the changes in frequency and energy at different time points. The crack signal shows a sudden high-frequency feature accompanied by a significant change in energy; the material fatigue signal shows a gradually developing low-frequency feature, and the energy change is relatively gentle;

[0019] Strain change graph generation: Generate a strain change graph based on the real-time data of the strain sensor to describe the strain distribution and change trend of the blade during operation. The signal of material fatigue is manifested as a gradual accumulation of strain, and the crack signal is accompanied by a sudden and drastic change in strain.

[0020] Double matching method: The double matching method achieves comprehensive distinction between crack signals and material fatigue signals by comparing and analyzing the acoustic feature time-frequency diagram with the strain change diagram.

[0021] Optionally, the double matching method specifically includes:

[0022] Frequency matching submodule: Analyze the frequency changes in the acoustic signal and match it with the strain change pattern in the material fatigue state. If the frequency changes are relatively gentle and conform to the trend of gradual strain accumulation, the signal is judged to be a material fatigue signal; if there is a sudden high-frequency change and the strain graph is accompanied by an obvious strain mutation, the signal is a crack signal;

[0023] Energy matching submodule: Analyzes the energy changes of acoustic signals. Material fatigue is usually manifested as a slow accumulation of energy, while crack signals are accompanied by a sudden release of energy. By comparing the synchronization of energy mutations and strain changes in acoustic signals, the type of abnormal acoustic signal is reconfirmed.

[0024] Optionally, the crack location module determines the crack location by combining data from multi-point acoustic sensors installed at different positions on the wind turbine blades. The acoustic sensor collects the relationship between the time difference of the crack signal and the distance the signal reaches each sensor. The distance between the crack location and each acoustic sensor is calculated through the geometric relationship between the time difference of the three points and the spatial coordinates. The coordinate position of the crack on the surface or inside the blade is calculated based on the trigonometric geometric positioning formula, thereby achieving precise positioning of the crack.

[0025] Beneficial effects of the present invention:

[0026] The present invention realizes real-time detection and positioning of cracks in wind turbine blades through the organic combination of an acoustic signal acquisition module, a material fatigue state monitoring module and a dynamic signal separation and identification module. In particular, the acoustic signal is decomposed at multiple levels through the Hilbert-Huang transform algorithm, and small frequency and energy changes are effectively identified, and normal acoustic signals are distinguished from abnormal signals. Combined with the data of multi-point sensors and a triangulation positioning algorithm, the crack positioning module can accurately determine the crack position, thereby solving the problems of insufficient crack detection accuracy and difficulty in positioning in the prior art, and significantly improving the accuracy and timeliness of crack detection.

[0027] In the present invention, the material fatigue state monitoring module monitors the strain, temperature and other changes of the material in real time through a multimodal sensor, compares it with the built-in material damage database, and outputs the material fatigue state data. On this basis, the dynamic signal separation and identification module uses a dual matching algorithm of fatigue state data and acoustic feature time-frequency diagram and strain change diagram to effectively distinguish crack signals from material fatigue signals, breaking through the limitations of traditional filtering algorithms. Through the coordinated analysis of material behavior and acoustic signals, the separation accuracy of abnormal signals is greatly improved, thereby more accurately detecting crack signals. BRIEF DESCRIPTION OF THE DRAWINGS

[0028] In order to more clearly illustrate the technical solutions in the present invention or the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings in the following description are only for the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0029] Figure 1 A schematic diagram of various functional modules of a system according to an embodiment of the present invention;

[0030] Figure 2 Schematic diagram of the double matching method according to an embodiment of the present invention. DETAILED DESCRIPTION

[0031] The present invention is described in detail below in conjunction with the accompanying drawings and specific embodiments. At the same time, it is explained here that in order to make the embodiments more detailed, the following embodiments are the best and preferred embodiments, and those skilled in the art may also adopt other alternatives to implement some known technologies; and the accompanying drawings are only for more specific description of the embodiments, and are not intended to specifically limit the present invention.

[0032] It should be noted that the references to "one embodiment", "an embodiment", "an exemplary embodiment", "some embodiments" and the like in the specification indicate that the embodiments described may include specific features, structures or characteristics, but not every embodiment may include the specific features, structures or characteristics. In addition, when a specific feature, structure or characteristic is described in conjunction with an embodiment, it should be within the knowledge of a person skilled in the art to implement such feature, structure or characteristic in conjunction with other embodiments (whether or not explicitly described).

[0033] In general, a term can be understood, at least in part, from its use in context. For example, depending, at least in part, on the context, the term "one or more" as used herein can be used to describe any feature, structure, or characteristic in the singular sense, or can be used to describe a combination of features, structures, or characteristics in the plural sense. Additionally, the term "based on" can be understood as not necessarily intended to convey an exclusive set of factors, but can instead, depending, at least in part, on the context, allow for the presence of other factors that are not necessarily explicitly described.

[0034] like Figure 1-Figure 2 As shown, the wind turbine blade crack detection system based on acoustic feature analysis includes an acoustic signal acquisition module, a material fatigue state monitoring module, a signal separation and identification module, and a crack positioning module, wherein:

[0035] The acoustic signal acquisition module installs acoustic sensors at multiple locations on the wind turbine blades to collect the acoustic signals generated during the operation of the blades in real time;

[0036] The material fatigue state monitoring module uses multimodal sensors to monitor the material property changes of blade materials in real time. The material property changes include stress changes and temperature gradients. The material fatigue state monitoring module has a built-in material damage database. It compares the material property changes with the material damage database and combines the historical operation data of the wind turbine blades to determine the material fatigue state of the current blades.

[0037] The signal separation and recognition module uses a nonlinear signal decomposition algorithm and a self-supervised learning model to perform multi-level analysis on the collected acoustic signals, and uses the Hilbert-Huang transform (HHT) to decompose the acoustic signals to distinguish between normal operating acoustic signals and abnormal acoustic signals. During the operation of wind turbine blades, normal acoustic signals have relatively stable frequency and energy distribution, while cracks or damage usually cause abnormal acoustic wave characteristics, such as irregular frequency changes or sudden energy changes. After HHT decomposition, these abnormal signals can be effectively identified and lay the foundation for subsequent crack signal identification and positioning. Abnormal acoustic signals also include crack signals and material fatigue signals. Based on the material fatigue state data output by the fatigue state monitoring module, the double matching method of the acoustic feature time-frequency diagram and the strain change diagram is used to distinguish the crack signal from the material fatigue signal. This module does not rely on traditional filtering algorithms, but performs intelligent separation through the differences in signal morphology and material behavior.

[0038] After distinguishing the crack signal, the crack location module combines the data of multi-point acoustic sensors and determines the crack location through a triangulation positioning algorithm.

[0039] The acoustic signal acquisition module installs array acoustic sensors at multiple predetermined positions on the wind turbine blades, including the root, middle and tip of the blades, to ensure full coverage of the key structural areas of the blades. The acoustic sensors are interconnected through a wireless communication network to form a distributed acoustic signal acquisition unit, which collects in real time the acoustic signals generated by the blades under different wind speeds, wind directions and load conditions. The collected acoustic signals include background acoustic signals during normal operation and abnormal acoustic signals that may be caused by cracks, fatigue damage or material deformation. The signal sampling frequency is adaptively adjusted according to the dynamic changes in the blade operating speed to ensure high-precision real-time data acquisition.

[0040] The material fatigue state monitoring module collects real-time data on stress changes and temperature gradient changes in the blades through a multimodal sensor array installed on the surface and inside of the blades. The multimodal sensor array includes strain sensors and thermal sensors, which capture the fatigue state of blade materials in multiple dimensions. The built-in material damage database of the material fatigue state monitoring module stores damage thresholds based on various blade material types and different operating environment conditions. By matching the real-time collected material property changes with the damage thresholds in the material damage database, and combining the operating history data of the wind turbine blades, the multivariate regression analysis method is used to evaluate the current material fatigue state of the blades in real time.

[0041] The strain ε is used to calculate the stress σ through the stress-strain relationship formula. The stress σ is used as a key parameter in the regression model together with other factors affecting the fatigue state to analyze the fatigue state of the material. The multivariate regression analysis method is expressed as:

[0042] Y=β 0 +β 1 σ+β 2 T g +∈, where Y is the fatigue state value, indicating the degree to which the material is close to fatigue damage, σ is the stress value calculated by the stress-strain relationship, T g is the temperature gradient, which indicates the local temperature change of the blade, collected by the thermal sensor, β 0 is a constant term, β 1 , β 2 is the regression coefficient obtained through historical data training, and ∈ is the error term.

[0043] In order to train the regression model with historical data and determine the regression coefficient β 0 ,β 1 ,β 2 , use the optimization algorithm to fit the model. The following are the specific training steps and calculation process:

[0044] Consider a historical dataset with the following variables:

[0045] σ: material stress under different conditions;

[0046] T g : Temperature gradient of the blade under different conditions;

[0047] Y: known material fatigue state (measured experimentally);

[0048] Each row of the dataset represents an observation under one experiment or run condition, and the format is as follows:

[0049] Where n is the number of samples of historical data.

[0050] 2. Regression equation: According to the previous regression model formula:

[0051] Y=β 0 +β 1 σ+β 2 T g +∈, we hope to determine the regression coefficient β through historical data 0 ,β 1 ,β 2 ;

[0052] 3. The optimization algorithm uses the least squares method, and the goal is to find a set of regression coefficients so that the actual observed fatigue state value Y i Fatigue state value predicted by the model The sum of squared errors between is minimized. The model predicts the value It can be expressed as:

[0053]

[0054] The goal is to minimize the following sum of squared errors:

[0055] Through the β 0 ,β 1 ,β 2 Take the derivative and set it to 0 to calculate the optimal regression coefficient value.

[0056] 4. Convert the regression problem into matrix form to solve it:

[0057] Let X be a matrix containing the independent variables:

[0058]

[0059] Y is the vector of target variables:

[0060]

[0061] β is the regression coefficient vector:

[0062]

[0063] The regression problem can be written as a matrix equation: Y = Xβ. To solve for β, the standard formula of the least squares method is used: β = (X T X) -1 X T Y, where X T is the transposed matrix of X, (X T X) -1 Yes X T The inverse matrix of X, X T Y is the product of X and Y. By using this formula, we can find the optimal β 0 ,β 1 and β 2 The value of .

[0064] The stress-strain relationship is calculated as: σ = E·ε, where E is the Young's modulus of the material and ε is the strain value measured by the strain sensor. The fatigue damage state is evaluated by calculating the fatigue state value Y through a multivariate regression model, which is then compared with the material damage threshold T damage For comparison: D fatigue is the evaluation value of the material fatigue damage state, when D fatigue ≥1, it means that the material has reached the critical point of fatigue. fatigue When <1, the material is in a safe state.

[0065] The signal separation and recognition module uses the empirical mode decomposition and Hilbert spectrum analysis in the Hilbert-Huang transform to decompose the acoustic signal in the time domain and frequency domain, extract each intrinsic mode function (IMF), identify the tiny frequency and energy change characteristics in the acoustic signal, analyze the frequency components and energy distribution of the decomposed signal, and then use the classifier to automatically distinguish between normally operating acoustic signals and abnormal acoustic signals.

[0066] (1) The role of empirical mode decomposition (EMD) is to decompose complex acoustic signals into a series of intrinsic mode functions (IMFs). These functions reflect the frequency changes of the signal at different time scales. Suppose the acoustic signal we collected is x(t). Through EMD decomposition, we can obtain multiple IMFs: Among them, IMF i (t) is the i-th intrinsic mode function, representing the essential oscillatory component of the signal, r(t) is the residual component, representing the low-frequency or long-term trend component, and n is the number of IMFs, which is automatically determined by the EMD decomposition process. Each IMF i (t) all reflect the signal components at different frequencies and time scales. By analyzing these IMFs i(t), identifying small frequency changes in the acoustic signal.

[0067] (2) Hilbert spectrum analysis: After obtaining each IMF i (t), the instantaneous frequency and energy distribution are calculated using the Hilbert transform, which is defined as follows:

[0068] Through Hilbert transform, we get the complex signal z i (t):

[0069] in;

[0070] is the instantaneous amplitude;

[0071] is the instantaneous phase.

[0072] By taking the derivative of the instantaneous phase, we can get the instantaneous frequency f i (t):

[0073]

[0074] By calculating each IMF i (t) is the instantaneous frequency f i (t) and instantaneous amplitude a i (t), construct the time-frequency spectrum of the acoustic signal, capture the tiny frequency and energy changes in the signal, and then distinguish normal from abnormal acoustic signals.

[0075] The classifier is trained in combination with a self-supervised learning model. The signal features (frequency and energy changes) obtained through Hilbert-Huang transform decomposition are input into the self-supervised learning model. The self-supervised learning model is trained through contrastive learning tasks. The model automatically learns how to distinguish between normal and abnormal signals without manually labeled data. Through training, a feature representation of the signal is generated, which is then used to automatically identify new signals and determine whether they are abnormal signals.

[0076] The self-supervised learning model is used to automatically distinguish normal acoustic signals from abnormal acoustic signals (such as crack signals and fatigue damage signals) from the decomposed signals. The signal features obtained by HHT decomposition are used as input data X to enter the self-supervised learning model;

[0077] Construct a self-supervised learning model f(X;θ), where X is the input signal feature and θ is the model parameter. The training process is as follows:

[0078] (1) Construct input data: The input data X includes the instantaneous frequency f of each IMF i (t), instantaneous amplitude a i(t) and other statistical features within the time period, let X be a matrix containing all signal features:

[0079] X=(f 1 (t) a 1 (t) … f n (t) a n (t));

[0080] (2) The self-supervised learning model performs unsupervised training by defining a contrastive learning task, automatically learning the feature representation of the signal. The loss function of the model can be defined as the contrastive learning loss: Among them, f(X i ; θ) is the model's response to the input signal X i The encoding, Yes X i Comparison sample after disturbance, τ is the temperature coefficient;

[0081] By minimizing the loss function L(θ), the characteristic representation of the signal is learned, which can effectively distinguish normal acoustic signals from abnormal acoustic signals.

[0082] (3) After the training is completed, the newly input acoustic signal is analyzed to identify whether it is an abnormal signal. The output result is a binary classification label, indicating whether the input signal is an abnormal acoustic signal: y = f(X; θ), where y represents the classification result, 1 represents an abnormal signal, and 0 represents a normal signal. Based on this classification result, the abnormal signal is further processed as a crack signal or a fatigue damage signal.

[0083] The double matching method of acoustic feature time-frequency diagram and strain change diagram is used to distinguish crack signals from material fatigue signals. The specific methods include:

[0084] Input of material fatigue state data: The fatigue state monitoring module outputs material fatigue state data. These data are based on the historical strain distribution, temperature gradient and matching results with the material damage database. The data reflect whether the material is in the fatigue stage, the severity of fatigue progression, and the health status of the material at a certain moment. These data are used for subsequent signal classification;

[0085] Acoustic feature time-frequency diagram generation: In the signal separation and identification module, after the Hilbert-Huang transform, the acoustic signal is decomposed into multiple intrinsic mode functions, and the instantaneous frequency and amplitude of the acoustic signal are extracted through Hilbert spectrum analysis. The acoustic feature time-frequency diagram is generated based on the extracted instantaneous frequency and amplitude of the acoustic signal, which shows the changes in frequency and energy at different time points. The crack signal shows a sudden high-frequency feature accompanied by a significant change in energy; the material fatigue signal shows a gradually developing low-frequency feature, and the energy change is relatively gentle;

[0086] Strain change graph generation: Generate a strain change graph based on the real-time data of the strain sensor to describe the strain distribution and change trend of the blade during operation. The signal of material fatigue is manifested as a gradual accumulation of strain, and the crack signal is accompanied by a sudden and drastic change in strain.

[0087] Double matching method: The double matching method achieves comprehensive distinction between crack signals and material fatigue signals by comparing and analyzing the acoustic feature time-frequency diagram with the strain change diagram.

[0088] The double matching method specifically includes:

[0089] Frequency matching submodule: Analyze the frequency changes in the acoustic signal and match it with the strain change pattern in the material fatigue state. If the frequency changes are relatively gentle and conform to the trend of gradual strain accumulation, the signal is judged to be a material fatigue signal; if there is a sudden high-frequency change and the strain graph is accompanied by an obvious strain mutation, the signal is a crack signal;

[0090] Energy matching submodule: Analyzes the energy changes of acoustic signals. Material fatigue is usually manifested as a slow accumulation of energy, while crack signals are accompanied by a sudden release of energy. By comparing the synchronization of energy mutations and strain changes in acoustic signals, the type of abnormal acoustic signals can be re-confirmed.

[0091] It also includes comprehensive decision-making: after completing the dual matching of frequency and energy, a comprehensive decision is made based on the output information of the fatigue status monitoring module:

[0092] If the fatigue state data indicates that the material is close to the critical point of fatigue damage, and the acoustic signal shows slow changes at low frequencies and steady energy accumulation, the system will determine that the abnormal acoustic signal is a material fatigue signal;

[0093] If the acoustic signal shows sudden high-frequency and high-energy changes, accompanied by drastic changes in strain, and the fatigue state data does not show that it is close to the critical point of damage, the system will determine that the abnormal acoustic signal is a crack signal;

[0094] Through the above double matching method, the abnormal acoustic signal can be further divided into a crack signal or a material fatigue signal, and the result is output to the subsequent module to determine the crack position.

[0095] The crack location module determines the crack location by combining the data from multi-point acoustic sensors installed at different positions on the wind turbine blades. The module calculates the relationship between the time difference of the crack signal collected by the acoustic sensor and the distance of the signal reaching each sensor. The module calculates the distance between the crack location and each acoustic sensor through the geometric relationship between the time difference of the three points and the spatial coordinates. The module also calculates the coordinate position of the crack on the surface or inside the blade based on the trigonometric geometric positioning formula, thereby achieving accurate positioning of the crack.

[0096] 1. Time difference measurement of multi-point acoustic sensors: Select three acoustic sensors at different positions of the blade, the positions of the sensors are P 1 (x 1 ,y 1 ,z 1 ),P 2 (x 2 ,y 2 ,z 2 ) and P 3 (x 3 ,y 3 ,z 3 ), the crack position is P c (x c ,y c ,z c ), the sound wave from the crack position P c The time for propagation to each sensor is t 1 ,t 2 and t 3 ;

[0097] Known conditions: The spatial coordinates P of the sensor 1 ,P 2 ,P 3 , the propagation speed of sound waves in the blade material is v;

[0098] 2. The distance formula of sound wave propagation: The distance of crack signal propagation has the following relationship with the time and speed of sound wave propagation: d 1 =v·t 1 ;d 2 =v·t 2 ;d 3 =v·t 3 ; where d 1 ,d 2 ,d 3 The sound waves are from the crack position P c To sensor P 1 ,P 2 ,P 3 distance;

[0099] 3. Time difference formula: Assume that the time differences collected by the sensor are Δt 12 =t 2 -t 1 and Δt 13 =t 3 -t 1 , the distance difference can be calculated according to the time difference formula:

[0100] Δd 12 =v·Δt 12 =d 2 -d1 ;

[0101] Δd 13 =v·Δt 13 =d 3 -d 1 ;

[0102] 4. According to the triangular geometry, the crack position P c (x c ,y c ,z c ) to the sensor is:

[0103]

[0104] Combined with the time difference formula Δd 12 and Δd 13 After calculation, we get the following two sets of equations:

[0105]

[0106] 5. Through the above two sets of equations, combined with the known sensor position P 1 (x 1 ,y 1 ,z 1 ),P 2 (x 2 ,y 2 ,z 2 ) and P 3 (x 3 ,y 3 ,z 3 ), the equations can be combined to solve the position coordinates of the crack P c (x c ,y c ,z c ), the equation is solved by an iterative method to finally determine the specific spatial position of the crack on the blade.

[0107] The present invention covers any substitution, modification, equivalent method and scheme made on the essence and scope of the present invention. In order to make the public have a thorough understanding of the present invention, specific details are described in detail in the following preferred embodiments of the present invention, but those skilled in the art can fully understand the present invention without the description of these details. In addition, in order to avoid unnecessary confusion about the essence of the present invention, well-known methods, processes, procedures, components and circuits are not described in detail.

[0108] The above is only a preferred embodiment of the present invention. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the principle of the present invention. These improvements and modifications should also be regarded as the scope of protection of the present invention.

Claims

1. A wind turbine blade crack detection system based on acoustic feature analysis, characterized in that: It includes acoustic signal acquisition module, material fatigue state monitoring module, signal separation and identification module, and crack location module, among which: The acoustic signal acquisition module collects the acoustic signals generated during the operation of the wind turbine blades in real time by installing acoustic sensors at multiple positions of the wind turbine blades; The material fatigue state monitoring module uses a multimodal sensor to monitor the material property changes of the blade material in real time. The material property changes include stress changes and temperature gradients. The material fatigue state monitoring module has a built-in material damage database, compares the material property changes with the material damage database, and combines the wind turbine blade operation history data to determine the material fatigue state of the current blade; The signal separation and recognition module performs multi-level analysis on the collected acoustic signals through a nonlinear signal decomposition algorithm and a self-supervised learning model, and decomposes the acoustic signals using the Hilbert-Huang transform to distinguish between normal operating acoustic signals and abnormal acoustic signals. The abnormal acoustic signals also include crack signals and material fatigue signals. Based on the material fatigue state data output by the fatigue state monitoring module, the double matching method of the acoustic feature time-frequency diagram and the strain change diagram is used to distinguish the crack signal from the material fatigue signal. After distinguishing the crack signal, the crack location module combines the data of the multi-point acoustic sensor and determines the crack position through a triangulation positioning algorithm.

2. The wind turbine blade crack detection system based on acoustic feature analysis according to claim 1 is characterized in that: The acoustic signal acquisition module installs array acoustic sensors at multiple predetermined positions on the wind turbine blades, including the root, middle section and tip of the blades. The acoustic sensors are interconnected through a wireless communication network to form a distributed acoustic signal acquisition unit, which collects the acoustic signals generated by the blades under different wind speeds, wind directions and load conditions in real time.

3. The wind turbine blade crack detection system based on acoustic feature analysis according to claim 1 is characterized in that: The material fatigue state monitoring module collects data on stress changes and temperature gradient changes of the blades in real time through a multimodal sensor array installed on the surface and inside of the blades. The multimodal sensor array includes strain sensors and thermal sensors to capture the fatigue state of the blade material in multiple dimensions. The built-in material damage database of the material fatigue state monitoring module stores damage thresholds based on various blade material types and different operating environment conditions. By matching the real-time collected material property changes with the damage thresholds in the material damage database, and combining the operating history data of the wind turbine blades, the multivariate regression analysis method is used to evaluate the current material fatigue state of the blades in real time.

4. The wind turbine blade crack detection system based on acoustic feature analysis according to claim 3 is characterized in that: The multivariate regression analysis method is expressed as: Y=β0+β1σ+β2T g +∈, where Y is the fatigue state value, indicating the degree to which the material is close to fatigue damage, σ is the stress value calculated by the stress-strain relationship, T g is the temperature gradient, which indicates the local temperature change of the blade, collected by the thermal sensor, β0 is the constant term, β1 and β2 are regression coefficients obtained by training with historical data, and ∈ is the error term.

5. The wind turbine blade crack detection system based on acoustic feature analysis according to claim 4 is characterized in that: The stress-strain relationship is calculated as: σ = E·ε, where E is the Young's modulus of the material, ε is the strain value measured by the strain sensor, and the fatigue damage state is evaluated by calculating the fatigue state value Y through a multivariate regression model, which is then compared with the material damage threshold T damage For comparison: D fatigue is the evaluation value of the material fatigue damage state, when D fatigue ≥1, it means that the material has reached the critical point of fatigue. fatigue When <1, the material is in a safe state.

6. The wind turbine blade crack detection system based on acoustic feature analysis according to claim 5 is characterized in that: The signal separation and identification module uses empirical mode decomposition and Hilbert spectrum analysis in the Hilbert-Huang transform to decompose the acoustic signal in the time domain and frequency domain, extract each intrinsic mode function, identify the tiny frequency and energy change characteristics in the acoustic signal, analyze the frequency components and energy distribution of the decomposed signal, and then use a classifier to automatically distinguish between normally operating acoustic signals and abnormal acoustic signals.

7. The wind turbine blade crack detection system based on acoustic feature analysis according to claim 6 is characterized in that: The classifier is trained in conjunction with a self-supervised learning model, and the signal features obtained by Hilbert-Huang transform decomposition are input into the self-supervised learning model. The self-supervised learning model is trained through a contrast learning task. The model automatically learns how to distinguish between normal and abnormal signals without manually labeled data. Through training, a feature representation of the signal is generated, which is then used to automatically identify new signals and determine whether they are abnormal signals.

8. The wind turbine blade crack detection system based on acoustic feature analysis according to claim 6, characterized in that: The dual matching method of acoustic feature time-frequency diagram and strain change diagram is used to distinguish crack signals from material fatigue signals, specifically including: Acoustic feature time-frequency diagram generation: In the signal separation and identification module, after the Hilbert-Huang transform, the acoustic signal is decomposed into multiple intrinsic mode functions, and the instantaneous frequency and amplitude of the acoustic signal are extracted through Hilbert spectrum analysis. The acoustic feature time-frequency diagram is generated based on the extracted instantaneous frequency and amplitude of the acoustic signal, which shows the changes in frequency and energy at different time points. The crack signal shows high-frequency characteristics accompanied by energy changes; the material fatigue signal shows gradually developing low-frequency characteristics, and the energy changes smoothly; Strain change graph generation: Generate a strain change graph based on the real-time data of the strain sensor to describe the strain distribution and change trend of the blade during operation. The signal of material fatigue is manifested as the gradual accumulation of strain, and the crack signal is accompanied by a dramatic change in strain. Double matching method: The double matching method comprehensively distinguishes crack signals and material fatigue signals by comparing and analyzing the acoustic feature time-frequency diagram with the strain change diagram.

9. The wind turbine blade crack detection system based on acoustic feature analysis according to claim 8, characterized in that: The double matching method specifically includes: Frequency matching submodule: Analyze the frequency changes in the acoustic signal and match it with the strain change pattern in the material fatigue state. If the frequency changes smoothly and conforms to the trend of gradual strain accumulation, the signal is judged to be a material fatigue signal; if high-frequency changes occur and the strain diagram is accompanied by a sudden change in strain, the signal is a crack signal. Energy matching submodule: Analyzes the energy changes of acoustic signals. Material fatigue is usually manifested as a slow accumulation of energy, while crack signals are accompanied by a sudden release of energy. By comparing the synchronization of energy mutations and strain changes in acoustic signals, the type of abnormal acoustic signal is reconfirmed.

10. The wind turbine blade crack detection system based on acoustic feature analysis according to claim 1, characterized in that: The crack location module determines the crack location in combination with data from multi-point acoustic sensors installed at different positions on the wind turbine blade. The acoustic sensor collects the relationship between the time difference of the crack signal and the distance of the signal reaching each sensor. The distance between the crack location and each acoustic sensor is calculated through the geometric relationship between the time difference of three points and the spatial coordinates, and the coordinate position of the crack on the surface or inside the blade is calculated based on the trigonometric geometric positioning formula.

Images