Fan blade strain monitoring method based on optical fiber sensing

Abstract

The invention discloses a fan blade strain monitoring method based on optical fiber sensing, and relates to the technical field of fan blade monitoring, and the method comprises the steps: collecting a vibration signal in a blade operation process, and carrying out the denoising and time-frequency analysis processing of the vibration signal; calculating a plurality of time-frequency parameters, carrying out frequency band division on the vibration signal by using wavelet packet transformation, and extracting a frequency index to obtain a signal characteristic parameter; establishing a multiple linear regression model between the blade strain and the signal characteristic parameters, constructing a health assessment model in combination with a stress analysis theory, and outputting a health assessment result; and evaluating the residual service life of the blade by combining the signal characteristic parameters with a health evaluation result according to a credibility analysis algorithm and a load subitem coefficient so as to carry out real-time monitoring on a strain state. The optical fiber sensors are arranged in the target monitoring area of the fan blade in the axial direction and the circumferential direction, the measurement principle of the optical time domain reflection technology is combined, and the spatial resolution and the measurement precision of strain monitoring are improved.

Description

Technical Field

[0001] The present invention relates to the technical field of wind turbine blade monitoring, and in particular to a wind turbine blade strain monitoring method based on optical fiber sensing. Background Art

[0002] As an indispensable component of clean energy, wind power plays an increasingly important role in the global energy mix. As a core component of wind turbine systems, the operating status of wind turbine blades is directly related to the performance, safety, and reliability of the entire system. However, due to their long-term exposure to harsh operating environments, wind turbine blades must withstand complex aerodynamic forces, gravity, and environmental factors such as strong winds, dust, and low temperatures, making them extremely susceptible to fatigue damage and crack propagation.

[0003] While various monitoring methods for wind turbine blades exist, they have numerous limitations. Traditional strain gauge-based monitoring methods require extensive wiring, which not only increases installation and maintenance costs but is also susceptible to interference and lacks stability in complex environments. While machine vision monitoring can provide some information on the blade surface, its effectiveness is easily affected by lighting and weather conditions, and it struggles to capture internal strain. While nondestructive testing techniques such as acoustic emission and ultrasonic testing can identify blade defects to a certain extent, these technologies cannot provide real-time and long-term monitoring of blade strain.

[0004] As wind power generation grows towards larger and deeper offshore wind turbines, turbine blades are increasingly larger. Traditional monitoring methods struggle to maintain the accuracy and coverage required for these large blades. Furthermore, existing monitoring methods often fail to fully consider multiple factors, such as material fatigue and load randomness, when assessing the remaining useful life of wind blades, resulting in inaccurate predictions.

[0005] Currently, no effective solutions have been proposed for the problems in related technologies. Summary of the Invention

[0006] In response to the problems in the related art, the present invention proposes a wind turbine blade strain monitoring method based on optical fiber sensing to overcome the above technical problems existing in the existing related art.

[0007] To this end, the specific technical solutions adopted in the present invention are as follows:

[0008] A method for monitoring fan blade strain based on optical fiber sensing, the method comprising:

[0009] S1. Collect vibration signals during blade operation and perform denoising and time-frequency analysis on the vibration signals;

[0010] S2. Based on the processing results, calculate various time-frequency parameters, and use wavelet packet transform to divide the vibration signal into frequency bands, extract frequency indicators to obtain signal characteristic parameters;

[0011] S3. Establish a multiple linear regression model between blade strain and signal characteristic parameters, and combine it with stress analysis theory to build a health assessment model and output the health assessment results;

[0012] S4. Combining the signal characteristic parameters with the health assessment results, the remaining service life of the blade is evaluated based on the credibility analysis algorithm and the load partial coefficient to perform real-time monitoring of the strain state.

[0013] Furthermore, collecting vibration signals during blade operation and performing denoising and time-frequency analysis on the vibration signals include:

[0014] S11. Synchronously collect blade vibration acceleration data using fiber optic sensors and local acceleration data sensors arranged along the axial and circumferential directions, and build a multi-dimensional early warning mechanism using digital twin technology.

[0015] S12. Based on the constructed multi-dimensional early warning mechanism, collect the vibration signals output by the optical fiber sensor and the acceleration data sensor, and use the wavelet filtering algorithm to denoise the vibration signals;

[0016] S13. Based on the denoised oscillation signal, a moving windowed fast Fourier transform algorithm is used to perform time-frequency analysis on the denoised vibration signal.

[0017] Furthermore, fiber optic sensors and local acceleration data sensors arranged along the axial and circumferential directions are used to synchronously collect blade vibration acceleration data. In combination with digital twin technology, a multi-dimensional early warning mechanism is constructed, including:

[0018] S111. Deploy optical fiber sensors based on optical time domain reflectometry technology in the target monitoring area of ​​the wind turbine blades along the axial and circumferential directions to collect stress distribution data of the blades during operation.

[0019] S112, using the deployed optical fiber sensor to emit pulsed light, and through the Rayleigh backscattering effect, to sense the local strain change caused by the blade vibration and identify the vibration event;

[0020] S113. Using optical time domain reflectometry, measure the round-trip time difference of the optical signal caused by the vibration event to determine the spatial location of the vibration event, and deploy an acceleration data sensor at the corresponding spatial location to synchronously collect acceleration data of the blade vibration;

[0021] S114. Use digital twin technology to establish a three-dimensional dynamic simulation model. Combined with Hooke's law, the stress and acceleration data are converted into a three-dimensional stress field. The dynamic driving equation is used to execute the real-time deformation response of the virtual blade under load.

[0022] S115. Based on the real-time deformation response results, the damage distribution of each unit is calculated, and a multi-dimensional early warning mechanism is constructed by combining the stress threshold and the root mean square value of the acceleration data.

[0023] Furthermore, the deployed fiber optic sensors emit pulsed light and, through the Rayleigh backscattering effect, sense the local strain changes caused by blade vibration. The vibration events identified include:

[0024] S1121. Divide the optical fiber into several sections along its length according to monitoring requirements;

[0025] S1122. Based on the division results, the optical fiber sensor emits pulsed light. During the propagation of the pulsed light, backward Rayleigh scattered light is generated. After superposition in each interval, the light returns to the incident end along the optical fiber, generating a detection signal.

[0026] S1123. Analyze the generated detection signal, analyze the amplitude vector sum and phase vector sum of the Rayleigh backscattered light in each interval, and calculate the scattered light intensity amplitude sum of each interval based on the amplitude and phase value of the Rayleigh backscattered light in the interval to obtain a light source signal.

[0027] S1124. Post-process the light source signal, combine the incident light power at the corresponding position with the optical fiber transmission loss, and extract the power change caused by the disturbance to identify the vibration event.

[0028] Furthermore, the expression of power change is:

[0029]

[0030] Where p B represents the power at point B which is d away from the fiber incident section; x represents the position variable from the fiber incident section; p o represents the initial power; d represents the distance from the optical fiber incident section; a0(x) represents the transmission loss of the incident light.

[0031] Furthermore, based on the constructed multi-dimensional early warning mechanism, the vibration signals output by the optical fiber sensor and the acceleration data sensor are collected, and the vibration signals are denoised using the wavelet filtering algorithm, including:

[0032] S121. Based on the Rayleigh backscattering effect, the optical fiber sensor emits pulsed light and detects the reflected signal to collect the changes in the optical fiber strain field caused by the blade vibration, obtains the corresponding optical signal, and converts the optical signal into an electrical signal.

[0033] S122. Based on the change in charge generated by the internal piezoelectric effect, the local acceleration data sensor collects acceleration data generated by the blade vibration and converts the acceleration data into an electrical signal.

[0034] S123, combining the electrical signal converted from the optical signal with the acceleration data to convert the electrical signal, establishing a vibration signal, and selecting a wavelet basis function and a number of decomposition layers to perform multi-layer decomposition on the vibration signal to obtain a smooth component and an oscillating component;

[0035] S124 , processing the wavelet coefficients in the oscillation component using a preset threshold, setting the wavelet coefficients smaller than the preset threshold to zero to remove noise interference, and reconstructing the processed wavelet coefficients to obtain a denoised vibration signal.

[0036] Furthermore, the time-frequency parameters include: mean, root mean square value, standard deviation, frequency domain energy, center of gravity frequency, and average frequency.

[0037] Furthermore, a multivariate linear regression model between blade strain and signal characteristic parameters is established, and combined with stress analysis theory, a health assessment model is constructed. The output health assessment results include:

[0038] S31. Establishing a multiple linear regression model between blade strain and signal characteristic parameters based on the acquired signal characteristic parameters;

[0039] S32. Combine stress analysis theory, Hooke's law and tensor analysis algorithm to obtain the stress components of the blade in different directions;

[0040] S33. Calculate the equivalent stress at each location based on the stress components in each direction, identify the stress concentration area, set the stress threshold, and evaluate the health status of the blade based on the change trend of the signal characteristic parameters over time;

[0041] S34. Based on the health status assessment result, a health assessment model based on a support vector machine is constructed, and the real-time signal characteristic parameters are input into the health assessment model to output the health assessment result of the current blade.

[0042] Furthermore, by combining the signal characteristic parameters with the health assessment results, the remaining service life of the blade is evaluated based on the credibility analysis algorithm and the load partial coefficient, so as to perform real-time monitoring of the strain state, including:

[0043] S41. Analyze the relationship between the signal characteristic parameters and the health assessment results, calculate the unit damage amount based on the damage accumulation theory, and calculate the residual strength of the blade in combination with the number of load cycles experienced by the blade;

[0044] S42. Based on the residual strength of the blade, a credibility analysis algorithm is used to calculate the credibility of the blade in the current state, and the credibility under multiple working conditions is obtained by combining the load partial coefficients under different working conditions;

[0045] S43. Evaluate the remaining service life of the blade under the current operating state by combining the reliability under multiple working conditions and the number of load cycles that have occurred.

[0046] Furthermore, the calculation formula of credibility is:

[0047]

[0048] Where R represents the reliability of the blade; σ y represents the standard deviation of blade strength; y represents the variable defined according to blade strength and working stress; μ y represents the y-mean of .

[0049] The beneficial effects of the present invention are:

[0050] 1. By deploying fiber optic sensors along the axial and circumferential directions in the target monitoring area of ​​the wind turbine blade, the present invention can achieve full coverage perception of strain in all directions of the blade. Combined with the measurement principle of optical time-domain reflectometry technology, the spatial resolution and measurement accuracy of strain monitoring are improved, and tiny strain changes can be captured more sensitively, thereby obtaining more comprehensive and accurate strain data.

[0051] 2. The present invention effectively removes noise interference in vibration data through wavelet filtering method, and combines time-frequency analysis technology to convert the signal into the time-frequency domain, so that the energy distribution characteristics of the signal at different times and frequencies can be clearly displayed. This not only lays a high-quality data foundation for the subsequent accurate extraction of strain characteristic parameters, but also improves the accuracy and stability of the overall data analysis.

[0052] 3. The present invention calculates multiple signal characteristic parameters from both the time domain and frequency domain dimensions, and introduces wavelet packet transform to achieve fine decomposition of vibration signals. It can comprehensively and accurately capture the vibration characteristics of wind turbine blades and accurately reflect their strain state, providing solid data support and judgment basis for in-depth analysis of blade structural behavior and assessment of their health status. BRIEF DESCRIPTION OF THE DRAWINGS

[0053] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0054] Figure 1is a flow chart of a method for monitoring strain of wind turbine blades based on optical fiber sensing according to an embodiment of the present invention;

[0055] Figure 2 The figure is a flow chart of a method for monitoring strain of wind turbine blades based on optical fiber sensing according to an embodiment of the present invention. DETAILED DESCRIPTION

[0056] To further illustrate each embodiment, the present invention provides drawings, which are part of the disclosure of the present invention. They are mainly used to illustrate the embodiments and can be used in conjunction with the relevant descriptions in the specification to explain the operating principles of the embodiments. By referring to these contents, ordinary technicians in this field should be able to understand other possible implementation methods and the advantages of the present invention.

[0057] According to an embodiment of the present invention, a method for monitoring wind turbine blade strain based on optical fiber sensing is provided.

[0058] The present invention will now be further described with reference to the accompanying drawings and specific embodiments. Figure 1 As shown, according to an embodiment of the present invention, a method for monitoring strain of wind turbine blades based on optical fiber sensing includes:

[0059] S1. Collect the vibration signal of the blade during operation, and perform denoising and time-frequency analysis on the vibration signal.

[0060] In this optional embodiment, collecting vibration signals during blade operation and performing denoising and time-frequency analysis on the vibration signals include:

[0061] S11. Through the fiber optic sensors and local acceleration data sensors arranged along the axial and circumferential directions, the acceleration data of the blade vibration is synchronously collected, and a multi-dimensional early warning mechanism is constructed in combination with digital twin technology.

[0062] In this optional embodiment, optical fiber sensors and local acceleration data sensors arranged along the axial and circumferential directions are used to synchronously collect acceleration data of blade vibration, and a multi-dimensional early warning mechanism is constructed in combination with digital twin technology, including:

[0063] S111. In the target monitoring area of ​​the wind turbine blade, optical fiber sensors based on optical time domain reflectometry technology are arranged in the axial and circumferential directions respectively to collect stress distribution data of the blade during operation.

[0064] S112. Utilize the deployed optical fiber sensors to emit pulsed light, and through the Rayleigh backscattering effect, sense the local strain changes caused by blade vibration and identify vibration events.

[0065] In this optional embodiment, pulsed light is emitted by deployed optical fiber sensors, and the local strain changes caused by blade vibration are sensed through the Rayleigh backscattering effect. Identifying vibration events includes:

[0066] S1121. Divide the optical fiber into several sections along its length according to monitoring requirements;

[0067] S1122. Based on the division results, the optical fiber sensor emits pulsed light. During the propagation of the pulsed light, backward Rayleigh scattered light is generated. After superposition in each interval, the light returns to the incident end along the optical fiber, generating a detection signal.

[0068] S1123. Analyze the generated detection signal, analyze the amplitude vector sum and phase vector sum of the Rayleigh backscattered light in each interval, and calculate the scattered light intensity amplitude sum of each interval based on the amplitude and phase value of the Rayleigh backscattered light in the interval to obtain a light source signal.

[0069] S1124. Post-process the light source signal, combine the incident light power at the corresponding position with the optical fiber transmission loss, and extract the power change caused by the disturbance to identify the vibration event.

[0070] In this optional embodiment, the expression of the power variation is:

[0071]

[0072] Where p B represents the power at point B which is d away from the fiber incident section; x represents the position variable from the fiber incident section; p o represents the initial power; d represents the distance from the optical fiber incident section; a0(x) represents the transmission loss of the incident light.

[0073] S113. Based on the optical time domain reflectometry technology, the round-trip time difference of the optical signal caused by the vibration event is measured to determine the spatial position of the vibration event. An acceleration data sensor is deployed at the corresponding spatial position to synchronously collect the acceleration data of the blade vibration.

[0074] S114. Use digital twin technology to establish a three-dimensional dynamic simulation model. Combined with Hooke's law, the stress data and acceleration data are converted into a three-dimensional stress field. The dynamic driving equation is used to execute the real-time deformation response of the virtual blade under load.

[0075] S115. Based on the real-time deformation response results, the damage distribution of each unit is calculated, and a multi-dimensional early warning mechanism is constructed by combining the stress threshold and the root mean square value of the acceleration data.

[0076] It should be noted that, assuming the total length of the optical fiber is L, the entire optical fiber is divided into N sections according to the accuracy, and the interval length is After the pulse light undergoes N scatterings, the Rayleigh scattered light generated will be superimposed at discrete locations and return to the incident point; let A k and is the amplitude vector sum and phase vector sum of M backscattered Rayleigh lights in the kth segment, a i and Ω i are the amplitude and phase values ​​of the i-th backward Rayleigh scattered light within ΔL, respectively. Then the amplitude and intensity of the scattered light of the k-th section of optical fiber can be expressed as:

[0077]

[0078] Where A k represents the amplitude vector sum of the backscattered Rayleigh light in the kth segment; j represents the imaginary unit; represents the phase vector sum of the backscattered Rayleigh light in the kth segment; M represents the amount of backscattered Rayleigh light; i represents the index value; a i represents the amplitude of the i-th backscattered Rayleigh light; Ω i represents the phase value of the i-th backscattered Rayleigh light.

[0079] Regarding the method of converting light intensity, let the power at point B, which is d away from the fiber incident section, be p B (x), then the power generated by the interference at this location in a short period of time is expressed as:

[0080]

[0081] Where p B (x) represents the power at point B, which is d away from the fiber incident segment; x represents the position variable from the fiber incident segment; p o represents the initial power; d represents the distance from the optical fiber incident section; a0(x) represents the transmission loss of the incident light.

[0082] To accurately locate vibration events, the distance from the event point to the incident segment can be determined by measuring the incident and return light information based on the principle of optical time domain reflectometry (OTDR):

[0083]

[0084] Where d is the distance from the optical fiber incident segment; c is the speed of light in a vacuum; τ is the time difference; and n is the refractive index.

[0085] At the same time, combined with the local acceleration sensor, when the sensor is subjected to vibration, the internal cantilever beam vibrates, causing the internal resistance to change, which in turn causes the charge Q to change. According to Newton's second law F=ma, it can be seen that Q is proportional to a, that is, Q=e*F=e*ma (e is the piezoelectric constant); finally, the charge is converted into a voltage signal through the charge amplifier, and a voltage value proportional to the acceleration is output, thereby obtaining the vibration acceleration information of the blade.

[0086] Then, digital twin technology is introduced to establish a three-dimensional dynamic simulation model, and a deep fusion driving system of fiber optic sensing data and virtual models is constructed; by developing a cross-scale mapping algorithm of strain-stress-modal, the axial strain ε of fiber optic monitoring is converted to x , hoop strain ε y The data are converted into a three-dimensional stress field using the formula derived from Hooke's law. The formula derived from Hooke's law is:

[0087]

[0088] Where σ ij represents the stress tensor at a certain point on the blade (i, j = 1, 2, 3, representing three mutually perpendicular directions respectively); E represents the elastic modulus of the material; v represents the Poisson's ratio of the material (describing the ratio of transverse strain to longitudinal strain); ε kk represents the sum of the strains in three directions (ε 11 +ε 22 +ε 33 );δ ij represents the Kronecker symbol; ε ij Represents the total strain at a certain point on the blade.

[0089] The real-time deformation deduction of the virtual blade is achieved through the dynamic driving equation. The unit damage distribution is dynamically calculated through the unit damage formula. At the same time, a multi-dimensional early warning mechanism is constructed based on the VonMises stress threshold and the root mean square value of the vibration acceleration. The dynamic driving equation is:

[0090]

[0091] Where M represents the blade mass matrix (related to mass distribution); represents the second-order derivative of the modal coordinate with respect to time; C represents the damping matrix (characterizing the vibration energy loss); represents the first-order derivative of the modal coordinate with respect to time; F(t) represents the external load (such as aerodynamic load, gravity); F sens (t) represents the load equivalent to the optical fiber sensing data (calculated by the sensor data equivalent load integration formula).

[0092] The equivalent load integration formula is:

[0093] F sens (t)=∫ Ω Β T σ(ε(t))dΩ;

[0094] Where, F sens (t) represents the load equivalent to the optical fiber sensing data (calculated by the integral formula); Ω represents the structural spatial integration domain covered by the optical fiber monitoring; B represents the unit strain-displacement matrix; T represents the matrix transpose; σ(ε(t)) represents the stress field calculated from the strain ε(t) through the constitutive relationship.

[0095] The unit damage formula is:

[0096]

[0097] Where D e represents the unit damage value (reflecting the degree of damage); k represents the unit number in the finite element model of the blade structure; i represents the index value of the load cycle, which is used to mark the i-th load cycle experienced by the blade; n i Represents the weight coefficient or quantity of the i-th unit (needs to be determined in combination with the specific modeling logic); N i represents the fatigue life under the stress level of the i-th unit; σ max,i It represents the maximum stress value borne by the kth element in the i-th load cycle.

[0098] The VonMises stress threshold is:

[0099]

[0100] Where σ VM represents the VonMises equivalent stress (also called equivalent yield stress); σ1, σ2, and σ3 represent the normal stresses in three orthogonal directions (shear-free stress state) obtained after eigenvalue decomposition of the stress tensor.

[0101] In addition, when the multi-dimensional early warning mechanism detects abnormal vibration signals in a certain area (such as the root mean square value of acceleration exceeding the set threshold, etc.), the system will mark the signal in this area as "high priority" and enable more stringent signal processing strategies (such as reducing noise tolerance, refining frequency discrimination, etc.) to ensure that key details in the abnormal signal are retained and effective information is avoided from being filtered out; when the early warning mechanism further triggers the "stress exceeding the standard" state, the model will focus on analyzing the changing trends of the signal characteristic parameters in this state (such as frequency domain energy mutation, frequency shift, etc.), and dynamically adjust the parameter weights in the evaluation model (such as increasing the proportion of stress concentration areas contributing to damage), thereby enhancing the sensitivity and accuracy of health status assessment.

[0102] S12. Based on the constructed multi-dimensional early warning mechanism, the vibration signals output by the optical fiber sensor and the acceleration data sensor are collected, and the vibration signals are denoised using the wavelet filtering algorithm.

[0103] In this optional embodiment, based on the constructed multi-dimensional early warning mechanism, collecting the vibration signals output by the optical fiber sensor and the acceleration data sensor, and using the wavelet filtering algorithm to denoise the vibration signals include:

[0104] S121. Based on the Rayleigh backscattering effect, the optical fiber sensor emits pulsed light and detects the reflected signal to collect the changes in the optical fiber strain field caused by the blade vibration, obtains the corresponding optical signal, and converts the optical signal into an electrical signal.

[0105] S122. Based on the change in charge generated by the internal piezoelectric effect, the local acceleration data sensor collects acceleration data generated by the blade vibration and converts the acceleration data into an electrical signal.

[0106] S123, combining the electrical signal converted from the optical signal with the acceleration data to convert the electrical signal, establishing a vibration signal, and selecting a wavelet basis function and a number of decomposition layers to perform multi-layer decomposition on the vibration signal to obtain a smooth component and an oscillating component;

[0107] S124 , processing the wavelet coefficients in the oscillation component using a preset threshold, setting the wavelet coefficients smaller than the preset threshold to zero to remove noise interference, and reconstructing the processed wavelet coefficients to obtain a denoised vibration signal.

[0108] S13. Based on the denoised oscillation signal, a moving windowed fast Fourier transform algorithm is used to perform time-frequency analysis on the denoised vibration signal.

[0109] It should be noted that the data of the backscattered Rayleigh light signal change caused by the change in the optical fiber refractive index due to blade vibration is collected, and the formula is as follows:

[0110]

[0111] Where p B (x) represents the power at point B, which is d away from the fiber incident segment; x represents the position variable from the fiber incident segment; p o represents the initial power; d represents the distance from the optical fiber incident section; a0(x) represents the transmission loss of the incident light; the collected optical signal is converted into electrical signal data related to the vibration.

[0112] The local acceleration sensor collects the acceleration data generated by blade vibration based on the principle of Q=e*F (Q represents the amount of internally generated charge, e represents the piezoelectric constant, F=ma, m represents mass, and a represents acceleration) and converts it into an electrical signal output.

[0113] First, the wavelet filtering method is used to denoise the data. The formula is:

[0114]

[0115] Where W f (a, b) represents the wavelet transform coefficients; f(t) represents the vibration signal; represents the conjugate of the wavelet basis function; a represents the scale parameter; b represents the translation parameter; t represents time.

[0116] After removing the noise, in order to obtain the characteristics of the signal more clearly, the following formula is used to perform time-frequency analysis on the denoised signal; STFT uses moving window fast Fourier transform calculation to obtain a series of signal spectra, and can reflect the change of the signal spectrum over time. The formula is as follows:

[0117]

[0118] Where STFT represents the moving windowed fast Fourier transform; j represents the imaginary unit; w represents the angular frequency; m represents the translation parameter of the window function w(tm); s(t) represents the vibration signal; and w(tm) represents the window function.

[0119] S2. Based on the processing results, multiple time-frequency parameters are calculated, and the vibration signal is divided into frequency bands using wavelet packet transform, and frequency indicators are extracted to obtain signal characteristic parameters.

[0120] In this optional embodiment, the time-frequency parameters include: mean value, root mean square value, standard deviation, frequency domain energy, center of gravity frequency, and average frequency.

[0121] It should be noted that, from the perspective of time domain, a variety of parameters that can reflect the signal amplitude and change characteristics are calculated; the mean is used as an indicator to describe the stable component of the signal, and the calculation formula is Among them, x i Represents the amplitude of the signal at the i-th sampling point, N represents the total number of sampling points; the root mean square value is used to measure the strength of the signal and its formula is The larger the value, the higher the overall energy of the signal, reflecting the severity of the fan blade vibration; the standard deviation is used to describe the degree of dispersion of the signal from the mean, and the formula is: The larger the standard deviation, the greater the fluctuation of the signal, which may mean that the strain state of the blade is unstable.

[0122] In the frequency domain, the fast Fourier transform (FFT) is used to convert the time domain signal to the frequency domain, and then the frequency domain energy is calculated; the calculation formula of the frequency domain energy is Among them, X(k) represents the frequency domain component after FFT transformation. This parameter reflects the energy distribution of the signal at different frequency components, which helps to analyze the frequency range where the blade vibration energy is concentrated. In addition, the maximum energy frequency f is determined. max , that is, f max =argmax k |X(k)|, which represents the frequency point where the energy in the signal is most concentrated.

[0123] Considering the complexity of the fan blade vibration signal, the present invention introduces wavelet packet transform to decompose the signal more finely; suppose the signal is decomposed into 2 n frequency bands, and the bandwidth of each frequency band is (f s represents the sampling frequency), the frequency band of the mth wavelet packet is:

[0124]

[0125] Where, fr m Indicates the frequency band of the mth wavelet packet; m represents the index value; f s Indicates the sampling frequency; 2 n+1 Indicates the number of frequency bands into which the signal is decomposed.

[0126] In order to further analyze the characteristics of the signal, the center frequency S1 and the average frequency S2 of the signal are calculated; the center frequency describes the frequency of the signal with the largest component in the spectrum, and its calculation formula is: Among them, P(k) represents the corresponding power spectrum value, f k Indicates the frequency amplitude of the corresponding point, which reflects the distribution center of the signal energy on the frequency axis; the average frequency is the average value of the power spectrum value, and the calculation formula is: This parameter reflects the average level of signal frequency from another perspective.

[0127] S3. Establish a multivariate linear regression model between blade strain and signal characteristic parameters, and combine it with stress analysis theory to construct a health assessment model and output the health assessment results.

[0128] In this optional embodiment, a multivariate linear regression model is established between blade strain and signal characteristic parameters, and combined with stress analysis theory, a health assessment model is constructed. The output health assessment results include:

[0129] S31. Establishing a multiple linear regression model between blade strain and signal characteristic parameters based on the acquired signal characteristic parameters;

[0130] S32. Combine stress analysis theory, Hooke's law and tensor analysis algorithm to obtain the stress components of the blade in different directions;

[0131] S33. Calculate the equivalent stress at each location based on the stress components in each direction, identify the stress concentration area, set the stress threshold, and evaluate the health status of the blade based on the change trend of the signal characteristic parameters over time;

[0132] S34. Based on the health status assessment result, a health assessment model based on a support vector machine is constructed, and the real-time signal characteristic parameters are input into the health assessment model to output the health assessment result of the current blade.

[0133] It should be noted that the correlation model between strain and characteristic parameters is first established; considering that in actual situations, the strain of wind turbine blades may be affected by a combination of factors, a multiple linear regression model is used to describe the relationship between strain and each characteristic parameter.

[0134] In order to more accurately evaluate the comprehensive strain state of the blade, stress analysis is introduced. According to Hooke's law, within the elastic range, there is a linear relationship between stress and strain σ = Eε (σ represents stress, E represents the elastic modulus of the material). For a complex structure like a wind turbine blade, considering its stress distribution in different directions, the tensor analysis method is used to describe the stress state. Let the stress tensor at a certain point of the blade be σ ij (i, j = 1, 2, 3, representing three mutually perpendicular directions respectively). By transforming and calculating the extracted strain characteristic parameters, the stress components of the point in different directions can be obtained.

[0135] In order to evaluate the overall stress condition of the blade, the concept of VonMises equivalent stress is introduced. For the plane stress state (simplified model, which can be expanded to three dimensions according to the actual situation), the calculation formula of VonMises stress is:

[0136]

[0137] Where, represents VonMises stress; σ 11 , σ 22 , σ 33 It represents the normal stress component of a certain point on the blade in three mutually perpendicular directions (usually defined as the x, y, and z axes), where σ 11 Corresponding to the normal stress in the x direction, σ 22 Corresponding to the normal stress in the y direction, σ 33 Corresponding to the normal stress in the z direction; σ 12 , σ 23 , σ 13 represents the shear stress component of a certain point on the blade in different direction planes, where σ 12 Corresponding to the shear stress in the xy plane, σ 23 Corresponding to the shear stress in the yz plane, σ 13Corresponding to the shear stress in the xz plane; in the plane stress, let σ 33 =σ 23 =σ 13 =0; By calculating the VonMises stress at different positions, the areas of stress concentration on the blade can be determined. These areas are often the areas prone to damage.

[0138] When evaluating the health of the blade, a stress threshold σ is set threshold ; When the calculated VonMises stress When the threshold is exceeded, it indicates that the part of the blade is in a high stress state and there is a potential risk of damage. At the same time, the health status of the blade can be further judged by combining the changing trend of the characteristic parameters over time.

[0139] S4. Combining the signal characteristic parameters with the health assessment results, the remaining service life of the blade is evaluated based on the credibility analysis algorithm and the load partial coefficient to perform real-time monitoring of the strain state.

[0140] In this optional embodiment, combining the signal characteristic parameters with the health assessment results, and based on the credibility analysis algorithm and the load partial coefficient, the remaining service life of the blade is assessed to perform real-time monitoring of the strain state, including:

[0141] S41. Analyze the relationship between the signal characteristic parameters and the health assessment results, calculate the unit damage amount based on the damage accumulation theory, and calculate the residual strength of the blade in combination with the number of load cycles experienced by the blade.

[0142] S42. Based on the residual strength of the blade, the credibility of the blade in the current state is calculated using a credibility analysis algorithm, and the credibility under multiple working conditions is obtained by combining the load partial coefficients under different working conditions.

[0143] In this optional embodiment, the calculation formula for credibility is:

[0144]

[0145] Where R represents the reliability of the blade; σ y represents the standard deviation of blade strength; y represents the variable defined according to blade strength and working stress; μ y represents the y-mean of .

[0146] S43. Evaluate the remaining service life of the blade under the current operating state by combining the reliability under multiple working conditions and the number of load cycles that have occurred.

[0147] It should be noted that according to Miner's damage accumulation theory, when a material is subjected to cyclic loads, its internal damage will gradually accumulate. When the damage accumulation reaches a certain level, the material will fail due to fatigue. Assume that the number of load cycles borne by the fan blade is N, the unit damage caused by each load cycle is D, and the initial strength of the blade is x o , then the residual strength x of the blade after N cycles N Expressed as:

[0148] x N =x0(1-ND);

[0149] Where x N Indicates the residual strength of the blade after N cycles; x o It represents the initial strength of the blade; N represents the number of load cycles that the wind turbine blade is subjected to; and D represents the unit damage caused by each load cycle.

[0150] In practical applications, determining the unit damage amount D is the key. By analyzing the extracted strain characteristic parameters and the calculated stress state, the unit damage amount D can be approximately expressed as:

[0151]

[0152] Where D represents the unit damage caused by each load cycle; N f It represents the fatigue life of a material under a specific stress amplitude.

[0153] Introducing the concept of credibility (reliability), assuming that the strength of the blade obeys the normal distribution Working stress follows normal distribution where μ s and σ s are the mean and standard deviation of the intensity, μ l and σ l are the mean and standard deviation of working stress respectively; define variable y=x s -x l (x s is the intensity, x l is the working stress), according to the credibility analysis algorithm (reliability principle), the credibility R of the blade is expressed as:

[0154]

[0155] Where R represents the reliability of the blade; σ y represents the standard deviation of blade strength; y represents the standard deviation of blade strength (σ s ) and working stress (σ l ) variables defined; μ y The y-mean of denoted by ; where μ y =μ s -μ l , μ s represents the mean value of blade strength; μ l represents the mean stress on the blade.

[0156] Considering that different types of loads have different effects on blade damage, the load partial coefficient γ is introduced. i , the credibility R1 under multiple working conditions is expressed as:

[0157]

[0158] Where R1 represents the reliability under multiple working conditions; σ y1 represents the standard deviation of variable y1; y represents the variable defined according to blade strength and working stress; μ y1 represents the mean of variable y1; y1 represents the variables involved in the reliability calculation formula under multiple working conditions; n represents the number of load cycles actually borne by the wind turbine blades; i represents the index value; γ i Indicates the weight coefficient of the i-th influencing factor; N i Indicates the number of load cycles that a material can withstand before fatigue failure occurs (i.e. fatigue life); D i represents the unit damage caused by each load cycle under the i-th working condition or stress level; x si represents a characteristic value of the blade strength under the i-th working condition; x li represents a characteristic value of the blade working stress under the i-th working condition; σ si represents the standard deviation of blade strength under the i-th working condition; σ li represents the standard deviation of the blade working stress under the i-th working condition; μ si represents the mean value of blade strength under the i-th working condition; μ li represents the mean value of the blade working stress under the i-th working condition.

[0159] The above method is used to calculate the life span N corresponding to the reliability of the blade in the current state. Given the number of load cycles N1 that has been experienced, the remaining service life T of the fan blade can be calculated. remaining for:

[0160]

[0161] Where, T remaining Indicates the remaining service life of the fan blade; N indicates the number of load cycles; N1 indicates the number of load cycles currently experienced; Indicates the average number of load cycles per unit time.

[0162] In this way, the strain state, stress level, material fatigue characteristics and randomness of the blades are comprehensively considered, and the remaining service life of the wind turbine blades can be predicted more accurately, providing a scientific basis for the maintenance and replacement of the wind turbine to ensure its safe and stable operation.

[0163] like Figure 2 As shown, in a specific embodiment, a method for monitoring wind turbine blade strain based on optical fiber sensing includes:

[0164] (1) Constructing a fiber optic sensing monitoring system;

[0165] (2) Collect and preprocess vibration data;

[0166] (3) Extracting strain characteristic parameters;

[0167] (4) Analyze strain status and assess health status;

[0168] (5) Predict the remaining useful life.

[0169] The (1) fiber optic sensing monitoring system is constructed, and fiber optic sensors are arranged along the axial and circumferential directions of the blades at key parts of the wind turbine blades, such as the root, middle and tip of the blades, which are prone to stress concentration. The axial arrangement is used to monitor axial strain, and the circumferential arrangement is used to monitor circumferential strain, so as to comprehensively obtain the strain information of the blades in different directions. A fiber optic sensor based on the optical time domain reflectometry (OTDR) principle is selected, and its working principle is based on the back Rayleigh scattering generated by the incident light in the optical fiber. When the optical fiber is subjected to external influences, such as the strain caused by the vibration of the wind turbine blades, the refractive index of the optical fiber will change, and then the intensity and phase of the back Rayleigh scattered light will change. According to the characteristics of the back Rayleigh scattered light, its light intensity is inversely proportional to the fourth power of the wavelength. By utilizing this relationship, the strain of the blades can be sensed by monitoring the changes in the back Rayleigh scattered light intensity.

[0170] In practical applications, in order to improve monitoring accuracy, improved OTDR technology is used, such as phase-sensitive optical time domain reflectometry Technology, by detecting and analyzing the phase of the backscattered Rayleigh light, can more sensitively monitor the tiny strain changes of the optical fiber; its basic principle is to process the light signal after interference and convert the phase change into the light intensity change for measurement. Let the total length of the optical fiber be L, and the entire optical fiber is divided into N sections according to the system accuracy, and the interval length is After the pulse light undergoes N scatterings, the Rayleigh scattered light generated will be superimposed at discrete locations and return to the incident point; let A k and is the amplitude vector sum and phase vector sum of M backscattered Rayleigh lights in the kth segment, a i and Ω iare the amplitude and phase values ​​of the i-th backward Rayleigh scattered light within ΔL, respectively. Then the amplitude and intensity of the scattered light of the k-th section of optical fiber can be expressed as:

[0171] Regarding the method of converting light intensity, the system receives the light source signal and performs post-processing such as bandpass and low-pass filtering and signal amplification to perform calculations; let the power at point B, which is d away from the fiber incident section, be p B (x), then the power generated by the interference at this location in a short period of time is expressed as:

[0172] To accurately locate vibration events, the distance from the event point to the incident section can be obtained by measuring the incident and return light information based on the principle of optical time domain reflectometry (OTDR):

[0173] At the same time, combined with local acceleration sensors, such as IEPE piezoelectric acceleration sensors, its working principle is based on the mutual conversion of force and acceleration; when the sensor is subjected to vibration, the internal cantilever beam vibrates, causing the internal resistance to change, which in turn causes the charge Q to change. According to Newton's second law F=ma, it can be seen that Q is proportional to a, that is, Q=e*F=e*ma (e is the piezoelectric constant); finally, the charge is converted into a voltage signal through a charge amplifier, and a voltage value proportional to the acceleration is output, thereby obtaining the vibration acceleration information of the blade.

[0174] Then, digital twin technology was introduced to establish a three-dimensional dynamic simulation model, and a deep fusion driving system of fiber optic sensing data and virtual models was constructed, breaking through the limitation of traditional monitoring based on analysis of only a single physical quantity; by developing a cross-scale mapping algorithm of strain-stress-modal, the axial strain ε of fiber optic monitoring was converted to x , hoop strain ε y The data are converted into a three-dimensional stress field using the formula derived from Hooke's law. The formula derived from Hooke's law is: The real-time deformation deduction of virtual blades is realized by dynamic driving equations. The present invention combines Miner damage accumulation theory with finite element fatigue analysis. The formula dynamically calculates the damage distribution of the element and is based on the VonMises stress threshold A multi-dimensional early warning mechanism is constructed with the root mean square value of vibration acceleration, so that the virtual model can not only reproduce the actual mechanical behavior of the blade, but also predict potential risk areas through parameter sensitivity analysis.

[0175] In the (2) collection and preprocessing of vibration data, after the optical fiber sensing monitoring system is constructed in (1), the system begins to continuously collect vibration data of the wind turbine blades during operation. The distributed optical fiber sensor collects the backscattered light signal change data generated by the change in the optical fiber refractive index caused by the blade vibration based on the principle of backscattered Rayleigh scattering of light. The formula is as follows:

[0176] The local acceleration sensor collects the acceleration data generated by blade vibration based on the principle of Q=e*F (Q represents the amount of internally generated charge, e represents the piezoelectric constant, F=ma, m represents mass, and a represents acceleration) and converts it into an electrical signal output.

[0177] The collected raw data often contains various noises, which will affect the accuracy of subsequent analysis, so preprocessing is required. First, the wavelet filtering method is used to denoise the data. The formula is: By selecting appropriate wavelet basis functions and decomposition layers, the collected vibration signal is decomposed into low-frequency (smooth component) and high-frequency (oscillating component) components. The low-frequency component mainly contains the main characteristic information of the signal, while the high-frequency component contains noise and detail information. The wavelet coefficients of the high-frequency component are processed by setting a threshold, and the wavelet coefficients smaller than the threshold are set to zero, thereby removing the noise. The processed wavelet coefficients are then reconstructed to obtain the denoised signal.

[0178] After removing the noise, in order to obtain the characteristics of the signal more clearly, the following formula is used to perform time-frequency analysis on the denoised signal; STFT uses moving window fast Fourier transform calculation to obtain a series of signal spectra, and can reflect the change of the signal spectrum over time. The formula is as follows: Through this transformation, the vibration signal in the time domain is converted to the time-frequency domain, and the energy distribution of the signal at different times and frequencies is obtained, so that the characteristics of the signal can be observed more intuitively, providing a better data basis for the subsequent accurate extraction of strain characteristic parameters.

[0179] In the (3) extraction of strain characteristic parameters, after completing (2) preprocessing the collected vibration data, characteristic parameters that can accurately characterize the strain state of the wind turbine blades are further extracted based on the signal after denoising and time-frequency analysis.

[0180] From the perspective of time domain, various parameters that can reflect the signal amplitude and change characteristics are calculated; the mean is used as an indicator to describe the stable component of the signal, and the calculation formula is: Among them, x i Represents the amplitude of the signal at the i-th sampling point, N represents the total number of sampling points; the root mean square value is used to measure the strength of the signal and its formula is The larger the value, the higher the overall energy of the signal, reflecting the severity of the fan blade vibration; the standard deviation is used to describe the degree of dispersion of the signal from the mean, and the formula is: The larger the standard deviation, the greater the fluctuation of the signal, which may mean that the strain state of the blade is unstable.

[0181] In the frequency domain, the fast Fourier transform (FFT) is used to convert the time domain signal to the frequency domain, and then the frequency domain energy is calculated; the calculation formula of the frequency domain energy is Among them, X(k) represents the frequency domain component after FFT transformation. This parameter reflects the energy distribution of the signal at different frequency components, which helps to analyze the frequency range where the blade vibration energy is concentrated. In addition, the maximum energy frequency f is determined. max , that is, f max =argmax k |X(k)|, which represents the frequency point where the energy in the signal is most concentrated.

[0182] Considering the complexity of the fan blade vibration signal, the present invention introduces wavelet packet transform to decompose the signal more finely; suppose the signal is decomposed into 2 n frequency bands, and the bandwidth of each frequency band is (f s represents the sampling frequency), the frequency band of the mth wavelet packet is:

[0183] In order to further analyze the characteristics of the signal, the center frequency S1 and the average frequency S2 of the signal are calculated; the center frequency describes the frequency of the signal with the largest component in the spectrum, and its calculation formula is: Among them, P(k) represents the corresponding power spectrum value, f k Indicates the frequency amplitude of the corresponding point, which reflects the distribution center of the signal energy on the frequency axis; the average frequency is the average value of the power spectrum value, and the calculation formula is: This parameter reflects the average level of signal frequency from another perspective.

[0184] In the above (4) analysis of strain state and assessment of health status, after completing (3) extraction of strain characteristic parameters, these parameters are used to deeply analyze the strain state of the wind turbine blades and assess their health status. Based on the extracted time domain and frequency domain characteristic parameters, a correlation model between strain and characteristic parameters is first established. Considering that in actual situations, the strain of wind turbine blades may be affected by a combination of multiple factors, a multivariate linear regression model is used to describe the relationship between strain and each characteristic parameter; let the strain value be ε, and the multiple characteristic parameters be x1, x2,…, x n (such as mean, root mean square value, frequency domain energy, etc.), establish the regression equation ε=β0+β1x1+β2x2+…++β n x n+ε, where β0,β1,…,β n is the regression coefficient, and ε is the error term. The regression coefficient is solved by the least squares method so that the model can best fit the relationship between strain and characteristic parameters.

[0185] In order to more accurately evaluate the comprehensive strain state of the blade, stress analysis is introduced. Referring to relevant mechanical theories, according to Hooke's law, within the elastic range, there is a linear relationship between stress and strain σ = Eε (σ represents stress, E represents the elastic modulus of the material). For a complex structure such as a wind turbine blade, considering its stress distribution in different directions, the tensor analysis method is used to describe the stress state. Let the stress tensor at a certain point of the blade be σ ij (i, j = 1, 2, 3, representing three mutually perpendicular directions respectively). By transforming and calculating the extracted strain characteristic parameters, the stress components of the point in different directions can be obtained.

[0186] In order to evaluate the overall stress condition of the blade, the concept of VonMises equivalent stress is introduced. For the plane stress state (simplified model, which can be expanded to three dimensions according to the actual situation), the calculation formula of VonMises stress is: By calculating the VonMises stress at different locations (under plane stress), the areas of stress concentration on the blade can be determined. These areas are often prone to damage.

[0187] When evaluating the health of the blade, a stress threshold σ is set threshold ; When the calculated VonMises stress When the threshold is exceeded, it indicates that the part of the blade is in a high stress state and there is a potential risk of damage. At the same time, the health status of the blade can be further judged by combining the changing trend of the characteristic parameters over time.

[0188] In addition, a large amount of historical monitoring data is studied and trained using machine learning algorithms to establish a health assessment model. The support vector machine (SVM) algorithm can be used to effectively classify the data in healthy and faulty states by finding an optimal classification hyperplane. During the training process, the extracted feature parameters are used as input, and the corresponding blade health status (healthy or faulty) is used as output. The parameters of the SVM are solved through the optimization algorithm so that it can accurately identify the health status of the blade. For new monitoring data, the extracted feature parameters are input into the trained SVM model, and the model will output the health assessment results of the blade, such as healthy, sub-healthy or faulty states.

[0189] In the above (5) prediction of the remaining service life, based on the completion of the strain state analysis and health status assessment of the wind turbine blades in (4), the remaining service life of the wind turbine blades is predicted by combining material fatigue theory and probability statistics methods.

[0190] First, according to Miner's damage accumulation theory, when a material is subjected to cyclic loads, its internal damage will gradually accumulate. When the damage accumulation reaches a certain level, the material will fail due to fatigue. Assume that the number of load cycles borne by the fan blade is N, the unit damage caused by each load cycle is D, and the initial strength of the blade is x. o , then the residual strength x of the blade after N cycles N Expressed as: x N =x0(1-ND).

[0191] In practical applications, determining the unit damage amount D is the key. D is estimated by analyzing the strain characteristic parameters extracted in (3) and the stress state calculated in (4), combined with the fatigue characteristic curve of the blade material (such as the SN curve, which describes the fatigue life of the material at different stress levels); according to the stress amplitude σ a ; Find the corresponding fatigue life N on the SN curve f , then the unit damage D can be approximately expressed as:

[0192] Considering the randomness and uncertainty of the load on the wind turbine blades in actual operation, a probability statistics method is used to more accurately predict the remaining service life; assuming that the strength of the blades obeys the normal distribution Working stress follows normal distribution where μ s and σ s are the mean and standard deviation of the intensity, μ l and σ l are the mean and standard deviation of working stress respectively; define variable y=x s -x l (x s is the intensity, x l is the working stress), according to the reliability principle, the reliability R of the blade is expressed as: in, μ y =μ s -μ l , through the statistical analysis of historical monitoring data, the mean and standard deviation of strength and working stress can be estimated; as the blade is used, its strength will gradually decrease, and according to Miner's damage accumulation theory, the mean and standard deviation of strength will also change; suppose that after N1 load cycles, the mean strength becomes μ s1 =(1-N1D)μ s, the standard deviation remains unchanged (this assumption can be made in the simplified model, and can actually be modified according to a more complex model). At this time, the N value corresponding to the reliability R (i.e., the life of the blade, expressed in the number of load cycles) can be obtained by solving the above integral equation through numerical calculation methods.

[0193] Furthermore, considering that different types of loads have different effects on blade damage, the load partial coefficient γ is introduced. i For wind turbine blades, common load types include aerodynamic loads, gravity loads, etc. Each load has a corresponding partial coefficient. The reliability R1 under multiple working conditions is expressed as:

[0194] The above method is used to calculate the life span N corresponding to the reliability of the blade in the current state. Given the number of load cycles N1 that has been experienced, the remaining service life T of the fan blade can be calculated. remaining for:

[0195] In this way, the strain state, stress level, material fatigue characteristics and randomness of the blades are comprehensively considered, and the remaining service life of the wind turbine blades can be predicted more accurately, providing a scientific basis for the maintenance and replacement of the wind turbine to ensure its safe and stable operation.

[0196] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. A method for monitoring fan blade strain based on optical fiber sensing, characterized in that: The method includes: S1. Collect vibration signals during blade operation and perform denoising and time-frequency analysis on the vibration signals; S2. Based on the processing results, calculate various time-frequency parameters, and use wavelet packet transform to divide the vibration signal into frequency bands, extract frequency indicators to obtain signal characteristic parameters; S3. Establish a multiple linear regression model between blade strain and signal characteristic parameters, and combine it with stress analysis theory to build a health assessment model and output the health assessment results; S4. Combining the signal characteristic parameters with the health assessment results, the remaining service life of the blade is evaluated based on the credibility analysis algorithm and the load partial coefficient to perform real-time monitoring of the strain state.

2. The method for monitoring wind turbine blade strain based on optical fiber sensing according to claim 1, characterized in that: The collecting of vibration signals during the operation of the blades and performing denoising and time-frequency analysis on the vibration signals include: S11. Synchronously collect blade vibration acceleration data using fiber optic sensors and local acceleration data sensors arranged along the axial and circumferential directions, and build a multi-dimensional early warning mechanism using digital twin technology. S12. Based on the constructed multi-dimensional early warning mechanism, collect the vibration signals output by the optical fiber sensor and the acceleration data sensor, and use the wavelet filtering algorithm to denoise the vibration signals; S13. Based on the denoised oscillation signal, a moving windowed fast Fourier transform algorithm is used to perform time-frequency analysis on the denoised vibration signal.

3. The method for monitoring fan blade strain based on optical fiber sensing according to claim 2, characterized in that: The optical fiber sensors and local acceleration data sensors arranged along the axial and circumferential directions are used to synchronously collect blade vibration acceleration data, and a multi-dimensional early warning mechanism is constructed in combination with digital twin technology, including: S111. Deploy optical fiber sensors based on optical time domain reflectometry technology in the target monitoring area of ​​the wind turbine blades along the axial and circumferential directions to collect stress distribution data of the blades during operation. S112, using the deployed optical fiber sensor to emit pulsed light, and through the Rayleigh backscattering effect, to sense the local strain change caused by the blade vibration and identify the vibration event; S113. Using optical time domain reflectometry, measure the round-trip time difference of the optical signal caused by the vibration event to determine the spatial location of the vibration event, and deploy an acceleration data sensor at the corresponding spatial location to synchronously collect acceleration data of the blade vibration; S114. Use digital twin technology to establish a three-dimensional dynamic simulation model. Combined with Hooke's law, the stress and acceleration data are converted into a three-dimensional stress field. The dynamic driving equation is used to execute the real-time deformation response of the virtual blade under load. S115. Based on the real-time deformation response results, the damage distribution of each unit is calculated, and a multi-dimensional early warning mechanism is constructed by combining the stress threshold and the root mean square value of the acceleration data.

4. The method for monitoring wind turbine blade strain based on optical fiber sensing according to claim 3, characterized in that: The method of using the deployed optical fiber sensor to emit pulsed light and sense the local strain changes caused by blade vibration through the backscattering effect to identify vibration events includes: S1121. Divide the optical fiber into several sections along its length according to monitoring requirements; S1122. Based on the division results, the optical fiber sensor emits pulsed light. During the propagation of the pulsed light, backward Rayleigh scattered light is generated. After superposition in each interval, the light returns to the incident end along the optical fiber, generating a detection signal. S1123. Analyze the generated detection signal, analyze the amplitude vector sum and phase vector sum of the Rayleigh backscattered light in each interval, and calculate the scattered light intensity amplitude sum of each interval based on the amplitude and phase value of the Rayleigh backscattered light in the interval to obtain a light source signal. S1124. Post-process the light source signal, combine the incident light power at the corresponding position with the optical fiber transmission loss, and extract the power change caused by the disturbance to identify the vibration event.

5. The method for monitoring wind turbine blade strain based on optical fiber sensing according to claim 4, characterized in that: The expression of the power variation is: Where p B represents the power at point B which is d away from the fiber incident section; x represents the position variable from the fiber incident section; p o represents the initial power; d represents the distance from the optical fiber incident section; a0(x) represents the transmission loss of the incident light.

6. The method for monitoring wind turbine blade strain based on optical fiber sensing according to claim 5, characterized in that: The multi-dimensional early warning mechanism based on the construction collects the vibration signals output by the optical fiber sensor and the acceleration data sensor, and uses the wavelet filtering algorithm to perform denoising on the vibration signals, including: S121. Based on the Rayleigh backscattering effect, the optical fiber sensor emits pulsed light and detects the reflected signal to collect the changes in the optical fiber strain field caused by the blade vibration, obtains the corresponding optical signal, and converts the optical signal into an electrical signal. S122. Based on the change in charge generated by the internal piezoelectric effect, the local acceleration data sensor collects acceleration data generated by the blade vibration and converts the acceleration data into an electrical signal. S123, combining the electrical signal converted from the optical signal with the acceleration data to convert the electrical signal, establishing a vibration signal, and selecting a wavelet basis function and a number of decomposition layers to perform multi-layer decomposition on the vibration signal to obtain a smooth component and an oscillating component; S124 , processing the wavelet coefficients in the oscillation component using a preset threshold, setting the wavelet coefficients smaller than the preset threshold to zero to remove noise interference, and reconstructing the processed wavelet coefficients to obtain a denoised vibration signal.

7. The method for monitoring wind turbine blade strain based on optical fiber sensing according to claim 1, characterized in that: The time-frequency parameters include: mean, root mean square value, standard deviation, frequency domain energy, center of gravity frequency, and average frequency.

8. The method for monitoring wind turbine blade strain based on optical fiber sensing according to claim 1, characterized in that: The multivariate linear regression model between blade strain and signal characteristic parameters is established, and combined with stress analysis theory, a health assessment model is constructed to output health assessment results including: S31. Establishing a multiple linear regression model between blade strain and signal characteristic parameters based on the acquired signal characteristic parameters; S32. Combine stress analysis theory, Hooke's law and tensor analysis algorithm to obtain the stress components of the blade in different directions; S33. Calculate the equivalent stress at each location based on the stress components in each direction, identify the stress concentration area, set the stress threshold, and evaluate the health status of the blade based on the change trend of the signal characteristic parameters over time; S34. Based on the health status assessment result, a health assessment model based on a support vector machine is constructed, and the real-time signal characteristic parameters are input into the health assessment model to output the health assessment result of the current blade.

9. The method for monitoring wind turbine blade strain based on optical fiber sensing according to claim 1, characterized in that: The method of combining the signal characteristic parameters with the health assessment results, evaluating the remaining service life of the blade based on the credibility analysis algorithm and the load partial coefficient, and performing real-time monitoring of the strain state includes: S41. Analyze the relationship between the signal characteristic parameters and the health assessment results, calculate the unit damage amount based on the damage accumulation theory, and calculate the residual strength of the blade in combination with the number of load cycles experienced by the blade; S42. Based on the residual strength of the blade, a credibility analysis algorithm is used to calculate the credibility of the blade in the current state, and the credibility under multiple working conditions is obtained by combining the load partial coefficients under different working conditions; S43. Evaluate the remaining service life of the blade under the current operating state by combining the reliability under multiple working conditions and the number of load cycles that have occurred.

10. The method for monitoring wind turbine blade strain based on optical fiber sensing according to claim 9, characterized in that: The calculation formula of the credibility is: Where R represents the reliability of the blade; σ y represents the standard deviation of blade strength; y represents the variable defined according to blade strength and working stress; μ y represents the y-mean of .

Images