Wind power blade internal defect detection method and system based on robot

Abstract

The invention relates to a wind power blade internal defect detection method and system based on a robot. The method comprises the following steps: acquiring and classifying wind power blade detection data to determine an abnormal type, dividing defect data associated with the abnormal type into a plurality of defect measurement points, defining an evaluation area based on the defect measurement points, and performing risk evaluation and risk verification on the evaluation area in a manner of performing weighted operation on the defect measurement points and comparing the defect measurement points with a risk threshold value. And outputting a risk assessment result. According to the method, the accuracy of defect measuring point identification is improved, the effective definition of the evaluation area is realized, and the risk of the evaluation area is objectively and quantitatively evaluated, so that reliable data support is provided for the maintenance decision of the wind power blade, and the operation and maintenance efficiency and safety are improved.

Description

Technical Field

[0001] This invention belongs to the field of wind turbine blade internal defect detection technology, specifically relating to a robot-based method and system for detecting internal defects in wind turbine blades. Background Technology

[0002] With the transformation of the global energy structure, wind power is a renewable energy technology. The blades in wind turbine generators are the core components for capturing wind energy. The structural integrity and operating status of the blades directly determine the power generation efficiency, operational stability and service life of the entire generator set. Therefore, internal defect detection of wind turbine blades is a key link to ensure the safe and efficient operation of wind farms.

[0003] Current technologies for detecting internal defects in wind turbine blades rely on manual visual inspection or handheld devices. This approach is not only labor-intensive and risky, but also inefficient, failing to meet the maintenance needs of large-scale wind farms. Furthermore, the results depend on the experience and subjective judgment of the inspectors, leading to inconsistent inspection standards, poor repeatability, missed detections, and misjudgments, posing potential safety hazards to the turbine. While existing automated inspection systems have improved efficiency to some extent, their data processing and analysis capabilities are limited. On the one hand, the complex environment during data acquisition results in low signal-to-noise ratios and a large amount of noise and artifacts in the raw inspection data. On the other hand, the back-end processing logic is relatively simple, often using fixed thresholds or simple pattern matching algorithms for defect judgment, lacking intelligent assessment and dynamic correction capabilities for data quality. This deficiency prevents the system from effectively distinguishing between genuine internal cracks and benign surface features, and also fails to identify and correct systematic data offsets caused by equipment vibration or inaccurate positioning during scanning, resulting in inaccurate defect location and size misjudgments.

[0004] To address the aforementioned issues, this application provides a robot-based method and system for detecting internal defects in wind turbine blades. Summary of the Invention

[0005] The purpose of this invention is to provide a robot-based method and system for detecting internal defects in wind turbine blades. This system can classify and process the detection data generated during the wind turbine blade inspection process to obtain the detection data and the corresponding internal defect classification. Based on the distribution trend of internal defects, multiple detection areas are divided, and the internal defect risk value of each detection area is evaluated. This effectively identifies the concentrated distribution area of ​​internal defects in the distribution process of internal defects in wind turbine blades and sets this area as a key detection area so as to focus on whether there are internal defects in such areas.

[0006] To achieve the above objectives, the technical solution adopted by the present invention is as follows: A robot-based method for detecting internal defects in wind turbine blades, comprising the following steps: Obtain wind turbine blade testing data. Identify one or more defect measurement points from wind turbine blade inspection data; Define the evaluation area based on one or more defect measurement points; The steps for defining the assessment area include: Collect the specific coordinate data of the defects at the defect measurement points; Obtain the blade width data of the blade to which the defect belongs; The evaluation zone is calculated based on specific coordinate data and blade width data; A risk assessment is conducted on the assessment area to generate a judgment result; Based on the judgment results, a risk verification is conducted on the assessment area to generate risk assessment results.

[0007] Preferably, the step of identifying one or more defect measurement points from wind turbine blade inspection data includes: The wind turbine blade inspection data is classified into inspection data that shows offset and inspection data that does not show offset; The anomaly type is determined based on the detection data that shows the shift and the detection data that does not show the shift. Defect data associated with anomaly types is divided into multiple defect measurement points.

[0008] Preferably, the step of classifying wind turbine blade test data includes: The change range of the current node's detection data is compared with a preset fluctuation range threshold to determine whether the wind turbine blade detection data shows a shift or does not.

[0009] Preferably, the step of determining the anomaly type based on the detection data showing offset and the detection data without offset includes: An alarm threshold is set based on the corrected offset detection data obtained after correcting the detection data that shows offset, and the detection data that does not show offset. Based on the alarm threshold, determine the anomaly type corresponding to the detected data that shows the offset.

[0010] Preferably, the steps for conducting a risk assessment of the assessment area include: Based on the calculation weights set for the defect measurement points, the data to be processed associated with the defect measurement points are calculated to generate calculation results; The calculation results are compared with the set risk thresholds to generate a judgment result indicating whether there are unconventional defects in the assessment area.

[0011] Preferably, the data to be processed includes defect elevation data of defect measurement points and their corresponding risk levels, extracted from a defect distribution group organized based on defect measurement points.

[0012] Preferably, the step of dividing the defect data into multiple defect measurement points includes: Defect data that are less than a preset distance threshold are grouped according to their spatial location to form defect groups; Each defect group is defined as a defect measurement point.

[0013] A robot-based system for detecting internal defects in wind turbine blades, comprising: The defect identification module is used to respond to the acquired wind turbine blade inspection data and identify one or more defect measurement points from the wind turbine blade inspection data; The evaluation area definition module is used to define the evaluation area based on one or more defect measurement points identified by the defect identification module. The risk assessment module is used to perform risk assessments on the assessment areas defined by the assessment area definition module in order to generate judgment results. The risk verification module is used to verify the defect risks within the assessment area in response to the judgment results generated by the risk assessment module, and output the risk assessment results.

[0014] Preferably, the defect identification module is configured to classify wind turbine blade inspection data into inspection data that exhibits offset and inspection data that does not exhibit offset; The anomaly type is determined based on the detection data that shows the shift and the detection data that does not show the shift. Defect data associated with anomaly types is divided into multiple defect measurement points.

[0015] Preferably, the evaluation area definition module is configured to: collect the specific coordinate data of defects in the defect measurement points; Obtain the blade width data of the blade to which the defect belongs; The evaluation zone is calculated based on specific coordinate data and blade width data; A risk assessment is conducted on the assessment area to generate a judgment result; Based on the judgment results, a risk verification is conducted on the assessment area to generate risk assessment results.

[0016] Beneficial effects This invention, based on preset classification rules, distinguishes raw detection data into detection data exhibiting deviation and detection data without deviation, and corrects the detection data exhibiting deviation. By applying a mathematical regression algorithm to the corrected deviation detection data and the detection data without deviation, an alarm trend curve is generated to dynamically set the alarm threshold. Through the classification, correction, and data-driven threshold setting method of the raw detection data, this invention can filter out signal noise and non-fault fluctuations during the detection process, and establish an objective and adaptive defect data identification standard, thereby improving the accuracy and reliability of identifying defect data from wind turbine blade detection data and avoiding the problems of missed or false alarms that may be caused by using fixed thresholds.

[0017] After determining the defect data, this invention associates it with physical defects inside the blade and collects the location and elevation data of the defects. Based on the spatial relationship of the defect data after arrangement, the defect data with a spatial distance less than a preset distance threshold are divided into defect measurement points and further organized into defect distribution groups. By integrating discrete defect points into defect measurement points with spatial clustering characteristics, this method can identify the defect concentration area inside the blade, providing a key basis for defining the evaluation area and assessing the health status of the local structure.

[0018] This invention uses defect measurement points as a basis, combines blade width data to calculate the evaluation area, and sets a calculation weight for each defect measurement point in the evaluation area. Based on the calculation weight, the defect elevation data of the defect measurement points are weighted to generate a quantitative calculation result, which is then compared with a risk threshold to determine whether there are any unconventional defects. This method establishes a multi-dimensional quantitative evaluation model that comprehensively considers the spatial distribution of defects, elevation position, blade geometry, and risk level. Attached Figure Description

[0019] Figure 1 This is a flowchart of the method in Embodiment 1 of the present invention. Detailed Implementation

[0020] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to specific embodiments. It should be understood that the specific embodiments described herein are merely for explaining the invention and are not intended to limit the scope of protection of the invention.

[0021] Example 1 This embodiment provides a robot-based method for detecting internal defects in wind turbine blades. By combining data filtering, feature recognition, classification processing, and defect risk assessment, it achieves refined monitoring of structural anomalies within wind turbine blades (e.g., structural discontinuities such as debonding, delamination, inclusions, voids, or cracks within the composite material structure of the blade). This method is applicable to automated inspection processes and on-site inspection tasks in wind turbine blade production lines, possessing high practicality and engineering scalability. The specific process is as follows: Acquire and classify wind turbine blade testing data.

[0022] The system acquires wind turbine blade inspection data, such as ultrasonic response values ​​or eddy current response values, collected by a robotic automated inspection platform along a preset path. To effectively distinguish between normal signal fluctuations and anomalies caused by potential defects, the system categorizes the wind turbine blade inspection data into those exhibiting offsets and those without offsets, based on a set of preset classification rules. Within a reference area where the blade structure is stable and has no historical defect records, one or more physical locations are selected as standard nodes. These standard nodes serve as the benchmark for subsequent signal comparisons to determine whether the signals at other inspection points are normal. Multiple historical inspection data points from various historical inspection tasks are collected for each standard node. By statistically calculating the variation amplitude of these historical data points, a numerical range is calculated. This range represents the normal variation amplitude of the inspection signal caused by equipment or environmental factors under defect-free conditions. A fluctuation range characterizing the normal signal variation at this node is determined, and a fluctuation amplitude threshold greater than the upper limit of the fluctuation range is set to determine the current inspection... The purpose of determining whether the signal change at the test node exceeds the critical value of the normal range is to establish a clear judgment boundary to filter out equipment noise or minor environmental disturbances. In actual testing, for each current node on the test path, the deviation of its test data relative to an evaluation baseline is calculated. The evaluation baseline is a reference signal value used to calculate the deviation of the current test node's signal. This baseline can be composed of the historical average signal value of a standard node or the theoretically designed signal value. This deviation is used as its amplitude. The amplitude of the change in the test data of the current node is compared with a preset fluctuation amplitude threshold. This fluctuation amplitude threshold is used as the offset judgment value. If the amplitude of change is less than the offset judgment value, the signal fluctuation is within the normal range, and the test data of the current node is determined to be test data without offset. Conversely, if the amplitude of change is greater than or equal to the offset judgment value, the signal is considered to have deviated significantly, and it is determined to be test data showing offset. This data is initially judged to be potentially related to defects.

[0023] Set alarm thresholds for the detection data that shows a shift and the detection data that does not show a shift.

[0024] For detection data classified as exhibiting offset, correction processing is required to eliminate or suppress known non-defect factor interference. For example, systematic deviations caused by changes in ambient temperature or equipment baseline drift can be subtracted to obtain corrected offset detection data. After ensuring the accuracy of subsequent analysis and completing data correction and classification, an alarm threshold needs to be set to achieve early warning of defect development trends. Specifically, all corrected offset detection data and non-offset detection data are plotted in the same two-dimensional coordinate system according to their spatial position on the blade, forming an original data curve that intuitively reflects the overall signal distribution. Since the evaluation curve may contain random noise points, direct use may lead to misjudgment. Therefore, mathematical fitting processing is required for the data points on the evaluation curve. For example, the least squares method or local weighted regression can be used to implement this fitting process, so as to generate a smoother alarm trend curve that reflects the true structural health trend. Based on the blade design specifications, material mechanical properties and historical failure data, a signal standard value representing an unacceptable structural state is pre-set and analyzed along the alarm trend curve to determine the node position on the curve where the signal value first exceeds the standard value. This node marks the critical point from acceptable to potentially dangerous structural state. Therefore, the signal value corresponding to this node is determined as the alarm threshold. This alarm threshold is not fixed, but dynamically generated according to the overall trend of the current batch of test data, which has higher adaptability and accuracy. The alarm threshold is a dynamically generated actual judgment basis for triggering defect alarms.

[0025] When setting alarm thresholds, the corrected offset detection data and the detection data without offset can be subjected to multiple mathematical fitting processes to establish one or more fitting curves. These curves reflect the response trend of the blade structure in a stable state. The inflection points in these curves that show significant changes can be used as candidates for alarm threshold nodes. Alarm threshold node: refers to the inflection point on the fitting curve where the signal trend changes significantly. This node is used as a candidate point to determine the final alarm threshold in order to avoid false triggering due to the oversensitivity of a single data point.

[0026] Calibrate data, associate defects, and categorize and archive them.

[0027] Detection data without deviation is labeled as normal data, indicating that the blade structure is in a healthy state at that location. Detection data showing deviation is labeled as fault data, indicating a potential anomaly in the blade structure at that location. Based on the dynamically set alarm threshold from the previous step, each fault data item is analyzed to determine its corresponding anomaly type. The anomaly type is classified according to the degree to which the fault data exceeds the alarm threshold, signal morphology, and other characteristics, classifying potential defects, such as minor anomalies, significant anomalies, or severe anomalies. Each fault data item is matched with its physical coordinates in the blade's 3D digital model, thereby associating the signal anomaly with a potential defect at a specific physical location inside the blade. This can manifest as structural discontinuities such as delamination layers, inclusion areas, hollow cavities, or crack bands inside the blade. This process includes collecting defect location data and defect elevation data. The defect elevation data is obtained by extracting the three-dimensional location data of the defect and the planar feature value associated with the defect fed back by the detection equipment (for example, in ultrasonic testing, this feature value is the round-trip propagation time of the ultrasonic wave). Based on the known propagation speed of the sound wave in the blade material and the measured propagation time, combined with the defect location data, the depth or height of the defect in the blade thickness direction can be calculated, which is the defect elevation data. After the parameter acquisition is completed, the defect location data and defect elevation data are labeled according to the determined anomaly type. The defect location data is used as the classification basis to determine the structural and functional area of ​​the defect in the blade, and the blade is divided into different areas. For example, it is classified into the root area where stress is more concentrated, the middle transition area, or the aerodynamically sensitive tip area, so as to assign differentiated importance weights to defects in different areas in subsequent risk assessment.

[0028] Divide the defect measurement points and generate defect distribution groups.

[0029] To assess the cumulative impact of defect clustering on structural safety, it is necessary to spatially integrate the discrete defects. Based on the annotations from the previous step, the location data and elevation data of defects with the same anomaly type are grouped into the same category, and these categorized location data and elevation data are uniformly defined as defect data. Traversing all defect data and based on their spatial coordinates, multiple defect data points with a spatial straight-line distance less than a preset threshold are grouped into a defect group, representing a spatially adjacent defect cluster. This grouping is necessary because the stress fields of multiple spatially adjacent small defects can superimpose, resulting in a combined effect far greater than that of a single defect; therefore, it is essential to treat them as a whole. Considering all factors, each defect group is defined as an independent defect measurement point. A defect measurement point is a logical unit defined to facilitate risk assessment. Each defect measurement point corresponds to a defect group, representing a defect cluster area that requires independent risk assessment. The risk level corresponding to each defect measurement point is obtained. This risk level can be initially determined based on the type and size of the defect by referring to a preset risk level comparison table. Based on the defect measurement points, the defect elevation data and corresponding risk level of each defect measurement point are compiled, and a structured defect distribution group containing information on multiple defect measurement points is output. The defect distribution group is a structured data set containing information on all identified defect measurement points for subsequent comprehensive risk assessment.

[0030] In the process of dividing defect measurement points, to address the issue of uneven defect data reading density, a method for calculating the probability of a node recurring in a specific area can be introduced to merge these nodes into a single logical measurement point. This simplifies the data and avoids redundancy. Multiple physical nodes with excessively high repetition rates and extremely close spatial locations are merged into a single virtual measurement point. By calculating the probability of a node recurring in a specific area and assigning a weight value that varies exponentially with a specific parameter (such as signal strength), the rationality and accuracy of the defect elevation data can be ensured, and the interpretability of subsequent risk assessment results can be improved.

[0031] The method for calculating the probability of a node recurring within a specific region is used to quantify the probability density of a defective node occurring at a specific spatial location. The calculation method is as follows:

[0032] In the formula, This represents the probability density, which means that at a given location... The probability of finding a defective node within a nearby unit of space; This represents the total number of nodes, which means the total number of defective node samples used for calculation. This indicates the evaluation location, which means the spatial coordinates of the target for which the probability density needs to be calculated; Indicates the position of a historical node, its meaning being the first... Spatial coordinates of historical defect nodes, set of location coordinates of historical defect nodes ,in or ; This represents bandwidth, which is a positive smoothing parameter that controls the smoothness of the probability density function and determines the influence range of neighboring nodes. Indicates the spatial dimension, which means the spatial dimension in which the node coordinates are located (for example, 2 in a two-dimensional plane and 3 in three-dimensional space). The kernel function is a non-negative function used to assign weights to neighboring points; for example, the Gaussian kernel function. .

[0033] Calculate the assessment area and risk weight.

[0034] After obtaining the defect distribution group, it is necessary to define specific physical areas for risk quantification. Specific coordinate data for each defect in the defect distribution group is collected. To accurately assess the relative severity of the defect on a specific blade cross-section, the blade width data of the blade to which the defect belongs also needs to be obtained. The blade width data is a calculated value representing the blade width at the cross-section where the defect is located, used to assess the relative positional importance of the defect. This data is precisely calculated as follows: Based on the three-dimensional coordinates of the plane where the defect is located, the blade profile of that plane is extracted from the blade's digital model. Two intersection points between the plane where the defect is located and the blade profile are determined. The straight-line distance between the two intersection points is calculated, yielding the intersection point difference, which is the basic value for calculating the blade width data. Considering that the structural contribution of the blade edge area is usually low, a preset ratio is used to correct the intersection point difference to obtain a corrected blade width that better reflects the width of the core load-bearing area. This corrected blade width is used as the blade width data. Based on the specific coordinate data of all defects and the calculated blade width data, the geometric boundary of the assessment area is defined by calculating the smallest bounding rectangle or convex hull that can contain the coordinates of all relevant defect measurement points. The assessment area is defined for risk quantification. The geometric region, whose boundary is determined by the smallest circumscribed geometric shape containing the coordinates of all relevant defect measurement points, such as a rectangle or convex hull; the dataset extracted from the evaluation area for risk weighting calculation, extracting defect elevation data and risk levels of defect measurement points to form the data to be processed, and assigning a calculation weight to each defect measurement point in the data to be processed. This weight comprehensively considers factors such as defect elevation and location. To achieve this, multiple numerical intervals are set as reference standards. For example, the defect elevation data is divided into multiple intervals, each corresponding to a different basic risk coefficient, and the data to be processed is weighted. The weighted calculation generates a single quantitative value by weighting the risks of all defect measurement points within the assessment area. This value comprehensively reflects the overall risk level of all defects within the assessment area. The specific calculation process is as follows: the risk level of each defect measurement point is weighted and summed or weighted averaged with its defect elevation data and the corrected blade width data at its location, thereby generating a quantitative calculation result that comprehensively reflects the overall risk level of all defects within the assessment area. This method ensures that the assessment of defects considers both their own severity and the structural criticality of their location.

[0035] Perform risk verification and output judgment.

[0036] The calculation results generated in the previous step are compared with a pre-set risk threshold based on a large amount of historical experimental data, simulation analysis, and engineering safety standards to generate a judgment result. The judgment result is a Boolean value or status code generated after comparing the calculation result with the risk threshold, which is used to directly indicate whether there are unconventional defects in the assessment area. The risk threshold represents the critical transition point from acceptable to unacceptable risk. Unconventional defects usually refer to those with extremely special size, density, or location, which, according to engineering experience, may lead to catastrophic structural failure. Based on the judgment result, a risk check is performed on the assessment area. If the judgment result indicates the presence of unconventional defects, the area is marked as the highest alarm level. If no unconventional defects are found, the assessment area is further risk-rated according to the specific values ​​of the calculation results and with reference to the preset risk classification standards, such as low, medium, and high risk. Finally, a final risk assessment result is output. This result is a comprehensive data report, which summarizes the precise coordinates, defect numbers, risk levels, related defect distribution group data, and corresponding detection signal segments of all identified high-risk areas and unconventional defect areas, providing comprehensive and quantitative data support for subsequent maintenance, reinforcement, or replacement decisions.

[0037] This embodiment can be further extended to the following technical features and method details: In the process of acquiring deviation data, in addition to conventional amplitude filtering, a grouping processing mechanism based on the spatial proximity of data points can be introduced. This mechanism automatically clusters continuous data points into groups using algorithms or rules, or uses a sorting method that dynamically adjusts the sequence of data values ​​to highlight outliers. Grouping analysis is performed on the data in the target area to automatically identify the trend bands of the signal. The trend bands are data segments that show a consistent trend of change within a certain spatial range, identified by the grouping processing mechanism. Positive and negative offset thresholds are set to capture abnormal data from two different directions: signal enhancement and signal attenuation. This creates a clear contrast with the stable region under normal conditions, thereby effectively reducing the false negative rate. The positive offset threshold is a specific value used to identify significant signal enhancement. When the signal value exceeds this threshold, it is judged as a positive anomaly. The negative offset threshold is a specific value used to identify significant signal attenuation. When the signal value is below this threshold, it is judged as a negative anomaly. The stable region is a stable range in which the numerical fluctuation of the detected signal remains between the positive and negative offset thresholds under normal conditions.

[0038] To correct the spatial position error of the defect elevation, the robot's position and orientation angles under different detection trajectories and the structural morphology data of the blade can be used. Combined with the offset between the coordinates of multiple measuring points and the original point, a calculation method based on geometric trigonometric relationships can be used to recalculate the defect elevation, so as to output a defect height value that is closer to the actual physical height of the structure.

[0039] A computational method based on geometric trigonometry is used to optimize the computational model of the defect's 3D coordinates by fusing observation data from multiple different robot poses to minimize reprojection error. The computational method is as follows:

[0040] In the formula, The optimized coordinates are the three-dimensional coordinates of the defect that are closest to its actual physical location, calculated after multi-view data fusion. This represents the coordinates to be optimized, which means the three-dimensional coordinate variables of the defect to be solved in the world coordinate system. This indicates the search for the variable that minimizes the objective function. ; The number of observations represents the total number of robot poses observed independently for the same defect. Represents the observation weight, its meaning is the first observation weight. The confidence weight of each observation can be set according to factors such as observation angle and signal-to-noise ratio. Indicates the observation position, its meaning is the first The location of the defect measured in the sensor's own coordinate system during the second observation; This represents the robot's pose, and its meaning is the first... At the time of the first observation, the homogeneous transformation matrix of the robot's end-sensor coordinate system relative to the world coordinate system, and the robot pose set. ,in For the first The 4x4 homogeneous transformation matrix for the second observation; the corresponding set of observed defect locations in the sensor coordinate system. ; This represents the projection function, which means projecting three-dimensional points in the world coordinate system. Transform to the The process of establishing a sensor coordinate system and then projecting it onto the sensor imaging plane can be described in this scheme if the sensor directly outputs three-dimensional coordinates. This is an identity transformation.

[0041] Before the final defect distribution group is formed, a key attribute weighting mechanism can be introduced. This mechanism converts the non-geometric attributes of defects, such as their shape characteristics, material properties at their location, and historical maintenance frequency, into weights by consulting a preset weight table or using function mapping. These weights are then integrated into the processing method of the risk calculation model to achieve the fusion calculation of multi-dimensional information.

[0042] In summary, this embodiment establishes a reasonable, well-defined, and highly interpretable technical solution for detecting internal defects in wind turbine blades through six steps: data acquisition, trend analysis, deviation identification, defect classification, distribution integration, and risk output. It is applicable to heterogeneous fusion detection scenarios involving multiple types of blades, multiple structural scenarios, and multiple types of defects.

[0043] Example 2 This embodiment provides a robot-based wind turbine blade internal defect detection system. This system executes the robot-based wind turbine blade internal defect detection method described in the preceding claims. Based on the wind turbine blade inspection data collected by the robot, it can automatically identify, evaluate, and verify potential defects inside the wind turbine blade, thereby achieving an assessment of the blade's health status. Specifically, this system can be a software program module executed by one or more processors, deployed on a server, industrial control computer, or cloud platform that communicates with the detection robot. The detection robot carries sensors such as ultrasonic, infrared thermal imaging, or acoustic emission sensors, moves inside the wind turbine blade, and collects data. Logically, the system can be divided into the following collaboratively working modules: Defect identification module The defect identification module is configured to respond to acquired wind turbine blade inspection data and identify one or more defect measurement points from the data. In a specific execution flow, it receives wind turbine blade inspection data transmitted in real-time or imported in batches by the inspection robot. It classifies the wind turbine blade inspection data by analyzing the inspection data at each current node in the data stream, comparing its variation amplitude with a preset fluctuation amplitude threshold. If the variation amplitude exceeds the threshold, the data is classified as inspection data showing an offset; otherwise, it is classified as inspection data without an offset. To determine the anomaly type, baseline correction or filtering is performed on the offset inspection data to obtain corrected offset inspection data. Based on this corrected offset detection... The system compares measured data with non-shifted detection data (as a reference for the normal state); sets one or more alarm thresholds; by comparing the original shifted detection data with these alarm thresholds, it can determine the anomaly type associated with the data point, such as delamination, debonding, or voids; identifies all data points associated with the identified anomaly types as defect data; to integrate discrete defect data into meaningful analysis units, these defect data are grouped according to spatial location; calculates the spatial straight-line distance between each defect data point; and groups all defect data with a spatial straight-line distance less than a preset distance threshold into the same defect group. Each defect group thus formed is ultimately defined as a defect measurement point and passed to subsequent modules for processing.

[0044] Evaluation area definition module The evaluation zone definition module is configured to define the evaluation zone based on one or more defect measurement points identified by the defect identification module. Upon receiving a defect measurement point, it collects and records the specific coordinate data of the defect contained in each defect measurement point, such as its three-dimensional position in the blade coordinate system. It accesses the pre-stored blade three-dimensional model database or design drawings to obtain the blade width data corresponding to the defect location, as well as other relevant structural dimension information (such as skin thickness and web position). Based on the collected specific coordinate data of the defect and the obtained blade width data, an evaluation zone is calculated and defined. The evaluation zone is a three-dimensional or two-dimensional spatial range surrounding the defect measurement point. Its size and shape are determined to cover the defect itself and the key structural areas that it may affect. For example, the evaluation zone can be defined as a rectangular or ellipsoidal area centered on the defect measurement point, with its boundary expanded proportionally according to the blade width and defect elevation data.

[0045] Risk assessment module The risk assessment module is configured to perform risk assessments on the assessment area defined by the assessment area definition module to generate judgment results. It processes all defect measurement points within the assessment area, organizing the defect measurement points and their related information into a defect distribution group. From this defect distribution group, it extracts the data to be processed for risk calculation. This data specifically includes the defect elevation data (i.e., the location of the defect in the blade thickness direction) of each defect measurement point and its corresponding risk level (which can be initially determined by the defect identification module based on the anomaly type). Calculation weights are assigned to different data to be processed; for example, defects near the blade pressure side surface or the load-bearing main beam are given higher calculation weights. The data to be processed is then weighted using these calculation weights to generate a comprehensive calculation result. This calculation result quantifies the risk index of the entire assessment area. This calculation result is compared with a pre-set risk threshold. If the calculation result is greater than or equal to the risk threshold, a judgment result indicating the possible existence of unconventional defects (i.e., high-risk defects) within the assessment area is generated; otherwise, a judgment result indicating that the risk is within an acceptable range is generated.

[0046] Risk verification module The risk verification module is configured to respond to the judgment results generated by the risk assessment module, perform risk verification on defects in the assessment area, and output the final risk assessment result; when the risk verification module receives a judgment result indicating the existence of non-routine defects, it initiates the risk verification process; The verification process may include: retrieving historical inspection data of the same location on the blade or blades of the same model for trend comparison analysis; matching the characteristics of the current defect (such as size, shape, and location) with the built-in expert knowledge base or fault tree model containing known defect cases; or automatically generating instructions to suggest a more precise re-inspection of the assessment area. The verification process is used to confirm the severity of the preliminary assessment, eliminate possible misjudgments, and conduct a more in-depth qualitative and quantitative analysis of the risk. After the verification is completed, the final risk assessment result is generated and output. The risk assessment result is a structured report that clearly indicates the specific location of the assessment area, the type and size of the defects contained therein, the confirmed risk level (e.g., "Level 1 Risk - Immediate shutdown for maintenance", "Level 2 Risk - To be addressed in planned maintenance", "Level 3 Risk - Continuous monitoring"), and corresponding maintenance recommendations.

[0047] Through the collaborative work of the aforementioned defect identification module, assessment area definition module, risk assessment module, and risk verification module, the system in this embodiment can automatically and continuously complete the entire process of detecting, locating, quantifying, assessing, and confirming risks of internal defects in wind turbine blades. It modularizes and automates the complex defect analysis process, improving detection efficiency and the reliability of risk assessment results. It is suitable for routine health monitoring and preventive maintenance scenarios of a large number of blades in large wind farms.

Claims

1. A robot-based method for detecting internal defects in wind turbine blades, characterized in that, The method includes: acquiring wind turbine blade testing data. Identify one or more defect measurement points from wind turbine blade inspection data; Define the evaluation area based on one or more defect measurement points; The steps for defining the assessment area include: Collect the specific coordinate data of the defects at the defect measurement points; Obtain the blade width data of the blade to which the defect belongs; The evaluation zone is calculated based on specific coordinate data and blade width data; A risk assessment is conducted on the assessment area to generate a judgment result; Based on the judgment results, a risk verification is conducted on the assessment area to generate risk assessment results.

2. The method for detecting internal defects in wind turbine blades based on robots according to claim 1, characterized in that, The steps for identifying one or more defect measurement points from wind turbine blade inspection data include: The wind turbine blade inspection data is classified into inspection data that shows offset and inspection data that does not show offset; The anomaly type is determined based on the detection data that shows the shift and the detection data that does not show the shift. Defect data associated with anomaly types is divided into multiple defect measurement points.

3. The method for detecting internal defects in wind turbine blades based on robots according to claim 2, characterized in that, The steps for classifying wind turbine blade inspection data include: The change range of the current node's detection data is compared with a preset fluctuation range threshold to determine whether the wind turbine blade detection data shows a shift or does not.

4. The method for detecting internal defects in wind turbine blades based on robots according to claim 2, characterized in that, The steps for determining the anomaly type based on the detection data with and without shift include: An alarm threshold is set based on the corrected offset detection data obtained after correcting the detection data that shows offset, and the detection data that does not show offset. Based on the alarm threshold, determine the anomaly type corresponding to the detected data that shows the offset.

5. The method for detecting internal defects in wind turbine blades based on robots according to claim 1, characterized in that, The steps for conducting a risk assessment of the assessment area include: Based on the calculation weights set for the defect measurement points, the data to be processed associated with the defect measurement points are calculated to generate calculation results; The calculation results are compared with the set risk thresholds to generate a judgment result indicating whether there are unconventional defects in the assessment area.

6. The method for detecting internal defects in wind turbine blades based on robots according to claim 5, characterized in that, The data to be processed includes defect elevation data and corresponding risk levels of defect measurement points extracted from defect distribution groups organized based on defect measurement points.

7. The method for detecting internal defects in wind turbine blades based on robots according to claim 2, characterized in that, The steps to divide defect data into multiple defect measurement points include: Defect data that are less than a preset distance threshold are grouped according to their spatial location to form defect groups; Each defect group is defined as a defect measurement point.

8. A robot-based system for detecting internal defects in wind turbine blades, characterized in that, include: The defect identification module is used to respond to the acquired wind turbine blade inspection data and identify one or more defect measurement points from the wind turbine blade inspection data; The evaluation area definition module is used to define the evaluation area based on one or more defect measurement points identified by the defect identification module. The risk assessment module is used to perform risk assessments on the assessment areas defined by the assessment area definition module in order to generate judgment results. The risk verification module is used to verify the defect risks within the assessment area in response to the judgment results generated by the risk assessment module, and output the risk assessment results.

9. A robot-based method for detecting internal defects in wind turbine blades according to claim 8, characterized in that, The defect identification module is configured to classify wind turbine blade inspection data into inspection data that shows offset and inspection data that does not show offset; The anomaly type is determined based on the detection data that shows the shift and the detection data that does not show the shift. Defect data associated with anomaly types is divided into multiple defect measurement points.

10. A robot-based method for detecting internal defects in wind turbine blades according to claim 8, characterized in that, The evaluation area definition module is configured to: collect the specific coordinate data of defects in the defect measurement points; Obtain the blade width data of the blade to which the defect belongs; The evaluation zone is calculated based on specific coordinate data and blade width data; A risk assessment is conducted on the assessment area to generate a judgment result; Based on the judgment results, a risk verification is conducted on the assessment area to generate risk assessment results.

Images