Wind turbine generator blade fault detection method based on multi-modal data

Abstract

The invention relates to the technical field of fault detection, and discloses a wind turbine generator blade fault detection method based on multi-modal data, and the method comprises the steps: obtaining the front edge line data of a blade and the preset power data of the blade, and generating a first power curve according to the preset power data; dividing the blade into a plurality of section sections based on the leading edge line data and a first power curve; calculating a real-time torque curve and real-time strain data of each section; comparing the real-time torque curve with a preset torque curve to obtain a first detection result; generating a second detection result according to the first detection result and the real-time strain data; and formulating a coping strategy according to the first detection result and the second detection result.

Description

Technical Field

[0001] This application relates to the field of fault detection technology, and in particular to a method for detecting faults in wind turbine blades based on multimodal data. Background Technology

[0002] Wind turbine blades are key components for capturing wind energy. Exposed to a complex and harsh environment for extended periods, they are highly susceptible to malfunctions such as leading-edge corrosion, surface cracks, lightning strike damage, internal structural delamination, and adhesive failure. If these malfunctions are not detected and addressed promptly, they can severely impact the turbine's power generation efficiency and even lead to blade breakage, resulting in significant economic losses.

[0003] Existing wind turbine blade fault detection methods suffer from three main shortcomings in achieving accurate diagnosis and efficient operation and maintenance. First, they lack the ability to differentiate fault types. When performance anomalies are detected, current technologies struggle to effectively distinguish the physical nature of the fault: they cannot determine whether it is surface damage affecting aerodynamic performance or internal damage affecting structural integrity; moreover, they cannot identify composite faults with both surface and internal damage. This lack of type differentiation leads to untargeted operation and maintenance decisions, significantly reducing maintenance efficiency and accuracy. Second, they are rigid in their response to minor or early-stage faults. Traditional methods rely heavily on global, fixed-threshold alarm mechanisms, failing to dynamically and meticulously assess the aerodynamic contribution and structural sensitivity of different blade sections. This results in insensitivity to minor faults and a susceptibility to false alarms due to fluctuations in operating conditions. Finally, fault location accuracy is limited. Most methods analyze the blade as a whole or in excessively large segments, leading to ambiguous fault location information and an inability to precisely pinpoint the problem to a specific blade section. This makes subsequent inspections and maintenance time-consuming and labor-intensive. These shortcomings collectively hinder the improvement of intelligent operation and maintenance in wind farms. Summary of the Invention

[0004] To address the aforementioned technical issues, this application provides a method for detecting wind turbine blade faults based on multimodal data, aiming to obtain a technical solution that can accurately locate fault positions and identify early, minor faults.

[0005] In some embodiments of this application, a method for detecting wind turbine blade faults based on multimodal data is disclosed, including:

[0006] Acquire blade leading edge data and blade preset power data, and generate a first power curve based on the preset power data;

[0007] Based on the leading edge data and the first power curve, the blade is divided into multiple cross-sectional segments;

[0008] Calculate the real-time torque curves and real-time strain data for each cross-section segment;

[0009] The real-time torque curve is compared with the preset torque curve to obtain the first detection result;

[0010] A second detection result is generated based on the first detection result and the real-time strain data;

[0011] Develop a response strategy based on the first and second test results.

[0012] In this embodiment of the application, generating the first power curve includes:

[0013] Generate the respective power curves of the blades at each position in one operating cycle based on the preset power data;

[0014] All the power curves are superimposed to generate the first power curve.

[0015] In this embodiment of the application, dividing the blade into multiple cross-sectional segments includes:

[0016] Obtain the first power curve of the blade;

[0017] The first power curve is integrated, and the cross-sectional segments are divided according to the integral value, wherein each cross-sectional segment satisfies the condition that the integral value is equal to a first threshold.

[0018] Obtain the curvature curve of the blade;

[0019] The integral value of the curvature curve for each divided cross-section is obtained sequentially;

[0020] For a segment whose curvature curve integral value is greater than the second threshold, the segment is further divided; for a segment whose curvature curve integral value is less than the second threshold, the segment is not further divided.

[0021] In this embodiment of the application, comparing the real-time torque curve with a preset torque curve to obtain a first detection result includes:

[0022] The difference torque curve of the cross-section is obtained by comparing the preset torque curve and the real-time torque curve at various positions of the cross-section within one operating cycle.

[0023] Set all the differential torque curves of a cross-section within one operating cycle as a differential torque curve group;

[0024] Set the differential torque curve set sequence C, C = (C1, C2…C…) i …C n ), where C1 is the differential torque curve set corresponding to the first cross-section segment, Y i This is the set of difference torque curves corresponding to the i-th cross-section segment. This represents the total number of cross-sectional segments;

[0025] Set the first and second conditions;

[0026] The cross-sectional segments that meet the first condition are marked as having a type of fault;

[0027] The cross-sectional segments that meet the second condition are marked as having type II faults;

[0028] The cross-section that simultaneously satisfies both the first and second conditions is marked as having three types of faults.

[0029] In this embodiment of the application, setting the first condition and the second condition includes:

[0030] The first condition is that, in the differential torque curve, the maximum torque value is greater than the first change threshold.

[0031] The second condition is that there exists a continuous region in the differential torque curve where all torques in the region are greater than the second variation threshold, and the span of this region along the blade length direction is greater than the preset length threshold.

[0032] In this embodiment of the application, generating a second detection result based on the first detection result and real-time strain data includes:

[0033] Let S be the sequence of cross-sections with three types of faults, S = (S1, S2, ..., Si, ..., Sn), where S1 is the first cross-section with three types of faults, Si is the i-th cross-section with three types of faults, and n is the total number of cross-sections with three types of faults.

[0034] Based on the differential torque curve, the location of the fault in each cross-section segment in the sequence of cross-section segments is obtained;

[0035] The location where the fault occurred is set as the fault location set;

[0036] Obtain real-time strain data of the blades at the fault location set;

[0037] The real-time strain data is compared with the preset strain data to generate a second detection result.

[0038] In this embodiment of the application, the step of comparing the real-time strain data with preset strain data to generate a second detection result includes:

[0039] Obtain historical strain data for each cross-section at each fault location in the cross-section segment sequence;

[0040] The preset strain data is obtained by averaging the historical strain data;

[0041] Acquire real-time strain data;

[0042] Calculate the absolute value of the difference between the historical strain data and the real-time strain data at each fault location for each cross section end;

[0043] If the absolute value is greater than the strain threshold, then the cross-sectional segment is classified as having four types of faults;

[0044] If the absolute value is less than the strain threshold, then the cross-sectional segment is set to not have the four types of faults.

[0045] In this embodiment of the application, the step of formulating a response strategy based on the first detection result and the second detection result includes:

[0046] Define a fault sequence G, G = (G1, G2, ..., Gi, ..., Gn), where G1 is the first fault on the blade, Gi is the i-th fault on the blade, and n is the total number of blade faults.

[0047] Calculate the severity score for each fault;

[0048] Different repair measures were taken for each fault.

[0049] In this embodiment of the application, the calculation of the severity score of each fault includes:

[0050] Obtain the location and type data for each fault;

[0051] Calculate the severity score F;

[0052] F = k1 * k2 * k3;

[0053] Where k1 represents the importance of the location; k2 represents the severity coefficient of the fault type; and k3 represents the difference between the preset torque and the real-time torque for each fault.

[0054] In this embodiment of the application, the different maintenance measures include:

[0055] When a Class I or Class II fault is detected, the image sensor on the blade is retrieved, and the fault condition on the blade surface is verified through image recognition.

[0056] When three types of faults are detected, the blades are marked as faulty blades and are manually repaired with maintenance tools during inspections.

[0057] When four types of faults are detected, the blades are marked as blades under maintenance, and personnel are arranged to carry out maintenance within the standard maintenance time.

[0058] Compared with existing technologies, the wind turbine blade fault detection method based on multimodal data in this application has the following advantages:

[0059] An intelligent blade health management mechanism integrating fault detection, type differentiation, and precise location was constructed. This method achieves accurate fault identification by fusing multimodal data such as aerodynamic torque and structural strain. During data analysis, the system not only identifies performance anomalies in the blades but also precisely locates the root cause of the fault through decoupled analysis of torque and strain data—when torque deviation is significant but strain is normal, it is identified as a surface aerodynamic fault; when strain deviation is significant but torque is normal, it is identified as internal structural damage; and when both are abnormal, it is identified as a composite fault. This precise fault classification capability transforms maintenance work from traditional general inspections to targeted treatment, greatly improving the focus and efficiency of maintenance.

[0060] Meanwhile, this method enables dynamic optimization of detection thresholds and flexible adjustment of the response mechanism. Based on the aerodynamic contribution, structural importance, and historical performance of different blade cross-sections, the system intelligently adjusts the detection thresholds and alarm priorities for each cross-section, ensuring that the detection strategy always maintains optimal alignment with the actual operating status of the blades and the wind farm's operation and maintenance goals. By dividing the blades into multiple functional segments and analyzing them independently, the system exhibits high sensitivity to minor, early-stage local faults, enabling precise fault location. This hierarchical and adaptive detection capability significantly improves the accuracy and timeliness of fault identification, providing a reliable guarantee for predictive maintenance and safe, economical operation of wind farms.

[0061] The plan also takes into account improving the quality of fault detection in order to save manpower and material resources and avoid waste of resources. Attached Figure Description

[0062] Figure 1 This application describes a wind turbine blade fault detection method based on multimodal data. Detailed Implementation

[0063] The specific embodiments of this application will be described in further detail below with reference to the accompanying drawings and examples. The following examples are used to illustrate this application, but are not intended to limit the scope of this application.

[0064] In the description of this application, it should be understood that the terms "center", "upper", "lower", "front", "rear", "left", "right", "vertical", "horizontal", "top", "bottom", "inner", "outer", etc., indicate the orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings. They are only for the convenience of describing this application and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, they should not be construed as limitations on this application.

[0065] The terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of technical features indicated. Therefore, a feature defined as "first" or "second" may explicitly or implicitly include one or more of that feature. In the description of this application, unless otherwise stated, "a plurality of" means two or more.

[0066] In the description of this application, it should be noted that, unless otherwise expressly specified and limited, the terms "installation," "connection," and "linking" should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral connection; they can refer to a mechanical connection or an electrical connection; they can refer to a direct connection or an indirect connection through an intermediate medium; and they can refer to the internal connection between two components. Those skilled in the art can understand the specific meaning of the above terms in this application based on the specific circumstances.

[0067] like Figure 1 As shown in the embodiment of this application, the wind turbine blade fault detection method based on multimodal data includes:

[0068] Acquire blade leading edge data and blade preset power data;

[0069] A first power curve is generated based on preset power data;

[0070] Based on the leading edge data and the first power curve, the blade is divided into multiple cross-sectional segments;

[0071] Calculate the real-time torque curves and real-time strain data for each cross-section segment;

[0072] The real-time torque curve is compared with the preset torque curve to obtain the first detection result;

[0073] A second detection result is generated based on the first detection result and the real-time strain data;

[0074] Develop a response strategy based on the results of the first and second tests.

[0075] Specifically, the leading edge line of a blade is the line formed by all the leading edge points of the blade, and it is also the outline of the front part of the blade that comes into contact with the airflow.

[0076] Specifically, the preset power data is obtained by using wind turbine blade load simulation software, such as HAWC2, OpenFAST, and GHBladed, to model the environment and wind turbine blades of that size. Based on blade element momentum theory, the wind turbine blades are discretized into multiple infinitely thin two-dimensional blade element units. The power data obtained through software simulation is the preset power data.

[0077] This model can be used to obtain torque, aerodynamic force, strain, and power data for each blade element at various positions.

[0078] Specifically, the first power curve is a rectangular coordinate system, with the horizontal axis representing the position of each blade element and the vertical axis representing the preset power data of each blade element. By stitching together the preset power data, a continuous curve can be formed.

[0079] Specifically, the cross-section is a segment of a leaf with a certain width, composed of multiple continuous leaf elements.

[0080] Specifically, the real-time torque curve is a rectangular coordinate system formed by connecting the real-time torque data of all leaf elements on the cross-section. The horizontal axis represents the position of each leaf element, the vertical axis represents the real-time torque data, and the origin represents the leaf element closest to the leaf root among all leaf elements on the cross-section.

[0081] Specifically, the preset torque curve is a rectangular coordinate system formed by connecting the preset torque data of all leaf elements on the cross section. The horizontal axis represents the position of each leaf element, the vertical axis represents the preset torque data, and the origin represents the leaf element closest to the leaf root among all leaf elements in the cross section.

[0082] The preset torque data is the same as the preset power data, obtained from the load simulation software.

[0083] In some embodiments of this application, generating a first power curve includes:

[0084] Generate the respective power curves of the blades at each position in one operating cycle based on the preset power data;

[0085] All power curves are superimposed to generate the first power curve.

[0086] Specifically, the top of the blade is set to 0 degrees and the bottom to 180 degrees. The rotation direction is clockwise. One operating cycle is when the blade rotates 360 degrees from 0 degrees back to the top. A data collection point is set at each degree, such as 0 degrees, 1 degree, 2 degrees...359 degrees. A total of 360 data collection points are set in one operating cycle.

[0087] Specifically, all power curves are superimposed to generate the first power curve, which is the power curve collected from 360 data acquisition points in one operating cycle from the preset power data, resulting in 36 power curves. The power data of each leaf element of all power curves are summed to obtain the final summed first power curve.

[0088] In some embodiments of this application, the blade is divided into multiple cross-sectional segments, including:

[0089] Obtain the first power curve of the blade;

[0090] Integrate the first power curve, and divide the cross-section according to the integral value. Each cross-section satisfies the condition that the integral value is equal to the first threshold.

[0091] Obtain the curvature curve of the blade;

[0092] The integral value of the curvature curve for each divided cross-section is obtained sequentially;

[0093] For a segment whose integral value of the curvature curve is greater than the second threshold, the segment is further divided; for a segment whose integral value of the curvature curve is less than the second threshold, the segment is not further divided.

[0094] Specifically, the first threshold is preferably 10.

[0095] Specifically, the divided cross-sectional segments are the individual cross-sectional segments obtained from the first division based on the power integral value.

[0096] Specifically, the second division means first dividing the cross-section into two cross-sections from the middle. If both cross-sections are less than the second threshold, the division ends. If not, the cross-section is divided into three new cross-sections from one-third to two-thirds of the original cross-section. This process continues until all newly divided cross-sections meet the requirement of being less than the second threshold.

[0097] Specifically, the first power curve is integrated, and cross-sectional segments are divided based on the integral value. Each cross-sectional segment satisfies the condition that the integral value is equal to the first threshold. Specifically, the integration is performed starting from the origin of the first power curve until the integral value is equal to the first threshold. This interval is then set as a cross-sectional segment, and the division continues until the entire blade is divided.

[0098] Specifically, the second threshold is preferably 3.

[0099] In some embodiments of this application, a first detection result is obtained by comparing the real-time torque curve with a preset torque curve, including:

[0100] The difference torque curve of the cross-section is obtained by comparing the preset torque curve and the real-time torque curve at various positions of the cross-section within one operating cycle.

[0101] Set all the differential torque curves of a cross-section within one operating cycle as a differential torque curve group;

[0102] Set the differential torque curve set sequence C, C = (C1, C2…C…) i …C n ), where C1 is the differential torque curve set corresponding to the first cross-section segment, Y i This is the set of difference torque curves corresponding to the i-th cross-section segment. This represents the total number of cross-sectional segments;

[0103] Set the first and second conditions;

[0104] The cross-sectional segments that meet the first condition are marked as having a type of fault;

[0105] The cross-sectional segments that meet the second condition are marked as having a type II fault;

[0106] The cross-section that simultaneously satisfies both the first and second conditions is marked as having three types of faults.

[0107] Specifically, the difference torque curve of the cross-section is obtained based on the preset torque curve and the real-time torque curve at each position of the cross-section within one operating cycle; specifically, the difference torque curve of all cross-sections at each data acquisition point within one operating cycle is generated.

[0108] Specifically, the differential torque curve set is a collection of 360 differential torque curves for a single cross-section during one operating cycle.

[0109] Specifically, the differential torque curve is obtained by subtracting the real-time torque data of the corresponding blade element from the preset torque data of each blade element on the cross-section. The curve formed by the differential torque data is the differential torque curve.

[0110] Specifically, a cross-sectional segment that meets the conditions is marked as having a corresponding fault. Specifically, if any differential torque curve in the differential torque curve group corresponding to the cross-sectional segment meets the conditions, it is determined that the cross-sectional segment has a corresponding type of fault.

[0111] Specifically, a cross-section that does not meet all the conditions is considered to be a cross-section that has not experienced a fault.

[0112] Specifically, the first type of fault is defined as a fault that occurs on the blade surface with a small area and a high degree of damage; the second type of fault is defined as a fault that occurs on the blade surface with a large area and a low degree of damage; and the third type of fault is defined as a fault that occurs deep within the blade or a fault with a large area or a high degree of damage.

[0113] In some embodiments of this application, the first condition and the second condition are set, including:

[0114] The first condition is that, in the differential torque curve, the maximum torque value is greater than the first change threshold.

[0115] The second condition is that there exists a continuous region in the differential torque curve where all torques in the region are greater than the second variation threshold, and the span of this region along the blade length direction is greater than the preset length threshold.

[0116] Specifically, real-time torque data is obtained by measuring the strain data of the cross section by attaching four strain gauges to the blade and connecting the four strain gauges to form a Wheatstone bridge.

[0117] First, calculate the relationship between the shear stress τ and torque T at a certain point on the surface: τ = T * r / J, where: T is the torque to be calculated; r is the radius of the blade at the measurement section; and J is the polar moment of inertia of the section. Then, according to the generalized Hooke's law, under pure shear stress, the maximum principal stress σ is equal to the shear stress τ. According to the generalized Hooke's law in mechanics of materials, the relationship between stress and strain is: σ = E * ε, where E is the elastic modulus of the material. Therefore, τ = E * ε. Combining these two equations, we obtain the torque calculation formula: T = (E * J * ε) / r, where: T = torque; E = elastic modulus of the blade composite material; J = polar moment of inertia of the measurement section; ε = net strain value measured by the strain gauge; and r = radius of the measurement section. Next, connect the output and strain: In a full-bridge circuit, the change in output voltage ΔV / V. V is the excitation voltage. The relationship between the voltage and strain ε is: ΔV / V = GF * ε / K; where: GF is the sensitivity coefficient of the strain gauge (provided by the manufacturer). K is the bridge coefficient; for a full bridge, K=4. The strain ε is calculated directly by measuring the output voltage of the bridge, and then substituted into the torque formula above. The torque formula is: T=(E*J) / r*((4*ΔV) / (GF*V)).

[0118] Specifically, the first variation threshold is the average value of the difference torque curves of each cross-sectional segment at each data acquisition point in the historical torque data. The second variation threshold is preferably one-fifth of the first variation threshold; the preset length threshold is set to half of the entire cross-sectional segment.

[0119] In some embodiments of this application, a second detection result is generated based on the first detection result and real-time strain data, including:

[0120] A second detection result is generated based on the first detection result and real-time strain data, including:

[0121] Let S be the sequence of cross-sections with three types of faults, S = (S1, S2, ..., Si, ..., Sn), where S1 is the first cross-section with three types of faults, Si is the i-th cross-section with three types of faults, and n is the total number of cross-sections with three types of faults.

[0122] Based on the differential torque curve, the location of the fault in each cross-section segment in the cross-section segment sequence is obtained;

[0123] Set the location of the fault as a fault location set;

[0124] Acquire real-time strain data of the blades at the fault location set;

[0125] The real-time strain data is compared with the preset strain data to generate a second detection result.

[0126] Specifically, the fault location set refers to the location points in the torque image that meet preset conditions and are judged as Class I, Class II, and Class III faults.

[0127] Specifically, the location of the fault in each cross-section in the cross-section sequence is obtained, and the location where the fault is detected is set as the location of the fault.

[0128] Specifically, real-time strain data is obtained by attaching fiber optic strain gauges or resistance strain gauges to the root of the blade. The strain gauges measure the minute length changes on the blade surface under load. The output signal is amplified and converted by an analog-to-digital converter to obtain the digital strain value.

[0129] In some embodiments of this application, comparing the real-time strain data with preset strain data to generate a second detection result includes:

[0130] Obtain historical strain data for each cross-section at each fault location in the cross-section segment sequence;

[0131] The preset strain data is obtained by averaging the historical strain data;

[0132] Acquire real-time strain data;

[0133] Calculate the absolute value of the difference between the historical strain data and the real-time strain data at each fault location for each cross section end;

[0134] If the absolute value is greater than the strain threshold, then the cross-sectional segment is classified as having four types of faults;

[0135] If the absolute value is less than the strain threshold, then the cross-sectional segment is set to not have the four types of faults.

[0136] Specifically, for each rotation of the blade, the strain sensor records a complete stress waveform. This generator superimposes and averages hundreds of waveforms to obtain a preset strain data. Then, it performs spectral analysis on this waveform and compares the actual data with this standard data to determine the internal fault of the blade.

[0137] Specifically, the preset strain data is obtained by averaging the strain data from multiple operating cycles.

[0138] In some embodiments of this application, a response strategy is formulated based on the first detection result and the second detection result, including:

[0139] Define a fault sequence G, G = (G1, G2, ..., Gi, ..., Gn), where G1 is the first fault on the blade, Gi is the i-th fault on the blade, and n is the total number of blade faults.

[0140] Calculate the severity score for each fault;

[0141] Different repair measures were taken for each fault.

[0142] Specifically, the severity score is based on the severity of each fault.

[0143] In some embodiments of this application, the severity score of each fault is calculated, including:

[0144] Obtain the location and type data for each fault;

[0145] Calculate the severity score F;

[0146] F = k1 * k2 * k3;

[0147] Where k1 represents the importance of the location; k2 represents the severity coefficient of the fault type; and k3 represents the difference between the preset torque and the real-time torque for each fault.

[0148] Classify the data in the dataset;

[0149] Based on data lineage, obtain the evaluation rules for the source data corresponding to each data point in the dataset;

[0150] Based on the data, select the model and set the rules for each data point in the overall data set;

[0151] Rules for the same type of data are processed to generate native data.

[0152] Specifically, the location data indicates the location of the fault. Different locations have different levels of importance. For example, a fault at the blade tip will have a greater impact on power generation, while a fault at the blade root will have a smaller impact.

[0153] Specifically, the importance of position is set by preset torque data, and the importance of position of each cross-section is equal to that of the other.

[0154] Specifically, the severity coefficient for each fault type is set as follows: 0.3 for Class I faults, 0.5 for Class II faults, 0.7 for Class III faults, and 0.9 for Class IV faults.

[0155] In some embodiments of this application, different maintenance measures are adopted, including:

[0156] When a Class I or Class II fault is detected, the image sensor on the blade is retrieved, and the fault condition on the blade surface is verified through image recognition.

[0157] When three types of faults are detected, the blades are marked as faulty blades and are manually repaired with maintenance tools during inspections.

[0158] When four types of faults are detected, the blades are marked as blades under maintenance, and personnel are arranged to carry out maintenance within the standard maintenance time.

[0159] Specifically, the standard repair time is preferably 24 hours.

[0160] The above are merely preferred embodiments of this application. It should be noted that, for those skilled in the art, several improvements and substitutions can be made without departing from the technical principles of this application, and these improvements and substitutions should also be considered within the scope of protection of this application.

Claims

1. A method for detecting wind turbine blade faults based on multimodal data, characterized in that, include: Acquire blade leading edge data and blade preset power data, and generate a first power curve based on the preset power data; Based on the leading edge data and the first power curve, the blade is divided into multiple cross-sectional segments; Calculate the real-time torque curves and real-time strain data for each cross-section segment; The real-time torque curve is compared with the preset torque curve to obtain the first detection result; A second detection result is generated based on the first detection result and the real-time strain data; Develop a response strategy based on the first and second test results.

2. The wind turbine blade fault detection method based on multimodal data as described in claim 1, characterized in that, The generation of the first power curve includes: Generate the respective power curves of the blades at each position in one operating cycle based on the preset power data; All the power curves are superimposed to generate the first power curve.

3. The wind turbine blade fault detection method based on multimodal data as described in claim 2, characterized in that, The process of dividing the blade into multiple cross-sectional segments includes: Obtain the first power curve of the blade; The first power curve is integrated, and the cross-sectional segments are divided according to the integral value, wherein each cross-sectional segment satisfies the condition that the integral value is equal to a first threshold. Obtain the curvature curve of the blade; The integral value of the curvature curve for each divided cross-section is obtained sequentially; For a segment whose curvature curve integral value is greater than the second threshold, the segment is further divided; for a segment whose curvature curve integral value is less than the second threshold, the segment is not further divided.

4. The wind turbine blade fault detection method based on multimodal data as described in claim 3, characterized in that, The step of comparing the real-time torque curve with a preset torque curve to obtain a first detection result includes: The difference torque curve of the cross-section is obtained by comparing the preset torque curve and the real-time torque curve at various positions of the cross-section within one operating cycle. Set all the differential torque curves of a cross-section within one operating cycle as a differential torque curve group; Set the differential torque curve set sequence C, C = (C1, C2…C…) i …C n ), where C1 is the differential torque curve set corresponding to the first cross-section segment, Y i This is the set of difference torque curves corresponding to the i-th cross-section segment. This represents the total number of cross-sectional segments; Set the first and second conditions; The cross-sectional segments that meet the first condition are marked as having a type of fault; The cross-sectional segments that meet the second condition are marked as having type II faults; The cross-section that simultaneously satisfies both the first and second conditions is marked as having three types of faults.

5. The wind turbine blade fault detection method based on multimodal data as described in claim 4, characterized in that, The setting of the first and second conditions includes: The first condition is that, in the differential torque curve, the maximum torque value is greater than the first change threshold. The second condition is that there exists a continuous region in the differential torque curve where all torques in the region are greater than the second variation threshold, and the span of this region along the blade length direction is greater than the preset length threshold.

6. The wind turbine blade fault detection method based on multimodal data as described in claim 5, characterized in that, The step of generating a second detection result based on the first detection result and real-time strain data includes: Let S be the sequence of cross-sections with three types of faults, S = (S1, S2, ..., Si, ..., Sn), where S1 is the first cross-section with three types of faults, Si is the i-th cross-section with three types of faults, and n is the total number of cross-sections with three types of faults. Based on the differential torque curve, the location of the fault in each cross-section segment in the sequence of cross-section segments is obtained; The location where the fault occurred is set as the fault location set; Obtain real-time strain data of the blades at the fault location set; The real-time strain data is compared with the preset strain data to generate a second detection result.

7. The wind turbine blade fault detection method based on multimodal data as described in claim 6, characterized in that, The step of comparing the real-time strain data with preset strain data to generate a second detection result includes: Obtain historical strain data for each cross-section at each fault location in the cross-section segment sequence; The preset strain data is obtained by averaging the historical strain data; Acquire real-time strain data; Calculate the absolute value of the difference between the historical strain data and the real-time strain data at each fault location for each cross section end; If the absolute value is greater than the strain threshold, then the cross-sectional segment is classified as having four types of faults; If the absolute value is less than the strain threshold, then the cross-sectional segment is set to not have the four types of faults.

8. The wind turbine blade fault detection method based on multimodal data as described in claim 7, characterized in that, The formulation of a response strategy based on the first and second detection results includes: Define a fault sequence G, G = (G1, G2, ..., Gi, ..., Gn), where G1 is the first fault on the blade, Gi is the i-th fault on the blade, and n is the total number of blade faults. Calculate the severity score for each fault; Different repair measures were taken for each fault.

9. The wind turbine blade fault detection method based on multimodal data as described in claim 8, characterized in that, The calculation of the severity score for each fault includes: Obtain the location and type data for each fault; Calculate the severity score F; F = k1 * k2 * k3; Where k1 represents the importance of the location; k2 represents the severity coefficient of the fault type; and k3 represents the difference between the preset torque and the real-time torque for each fault.

10. The wind turbine blade fault detection method based on multimodal data as described in claim 8, characterized in that, The different maintenance measures adopted include: When a Class I or Class II fault is detected, the image sensor on the blade is retrieved, and the fault condition on the blade surface is verified through image recognition. When three types of faults are detected, the blades are marked as faulty blades and are manually repaired with maintenance tools during inspections. When four types of faults are detected, the blades are marked as blades under maintenance, and personnel are arranged to carry out maintenance within the standard maintenance time.

Images