Defect area automatic identification method applied to terahertz transmission imaging

Abstract

The invention discloses a defect area automatic identification method applied to terahertz transmission imaging, and relates to the technical field of wind power generation blade defect detection and identification, and the method comprises the steps: firstly guaranteeing the detection coverage completeness through early-stage cooperative calibration and path planning, then obtaining high-quality original data through background collection and terahertz transmission imaging collection, and finally obtaining a detection result. Data quality is optimized through preprocessing such as background removal and noise reduction, suspected defect areas are marked based on pixel response values, accurate defect recognition is completed in combination with a trained model, and meanwhile detection precision and practicability are further improved through optimization means such as area grading scanning, temperature and humidity and attitude compensation and defect trend analysis. According to the method, the problems of weak anti-interference capability, insufficient precision, poor adaptability and lack of trend pre-judgment in a traditional detection method are effectively solved, full-process automation from data acquisition to result output is realized, and technical support is provided for efficient maintenance of the wind power generation blade.

Description

Technical Field

[0001] This invention relates to the field of wind turbine blade defect detection and identification technology, specifically to an automatic defect region identification method applied to terahertz transmission imaging. Background Technology

[0002] As an important component of clean energy, wind power is gradually increasing its share in the energy structure. Wind turbine blades are the core components of wind turbines, and their operating status directly affects power generation efficiency and equipment safety. During long-term use, blades are susceptible to defects such as surface cracks, internal delamination, and material aging due to environmental factors (such as wind, rain, dust, and temperature changes) and operating load. If these defects are not detected and addressed in a timely manner, they may lead to serious safety accidents such as blade breakage.

[0003] Current wind turbine blade defect detection technology has the following problems:

[0004] Firstly, the structural complexity, load-bearing requirements, and damage probability vary greatly in different areas of the blade. Fixed scanning parameters either cause redundancy in accuracy and reduce efficiency in conventional areas, or insufficient accuracy in key areas, leading to missed detections. Traditional data splicing is prone to splicing traces due to coordinate deviations and signal inconsistencies, which may cover up or misjudge defects.

[0005] Secondly, outdoor testing is subject to large temperature fluctuations, and terahertz signals are sensitive to temperature, which can easily cause amplitude drift and lead to distortion of defect characteristics. Long-term operation of equipment results in aging and wear, which further aggravates signal deviation. Traditional testing does not take these factors into account, and the test results are greatly affected by the environment and equipment condition, resulting in poor stability.

[0006] Third, during flight, airflow interference can cause changes in the distance between the probe and the blade, as well as attitude shifts, leading to inaccurate focusing of terahertz waves and changes in the incident angle, resulting in signal amplitude attenuation and phase distortion. Traditional imaging does not compensate for attitude changes, resulting in inconsistent detection accuracy at different parts of the curved blade and large defect identification errors.

[0007] Fourth, traditional inspection can only determine the current defect status and cannot predict the defect development trend. Maintenance work is mostly remedial after the fact, making it difficult to prevent risks in advance. There is a lack of basis for dividing maintenance priorities, and maintenance personnel cannot allocate resources reasonably, resulting in untimely handling of defects in critical areas and excessive maintenance in non-critical areas.

[0008] To address the aforementioned technical shortcomings, a solution is proposed. Summary of the Invention

[0009] The purpose of this invention is to solve the problems mentioned above by proposing an automatic defect region identification method for terahertz transmission imaging.

[0010] The objective of this invention can be achieved through the following technical solutions:

[0011] An automatic defect region identification method applied to terahertz transmission imaging is described, and the steps of the automatic defect region identification method are as follows:

[0012] Step S1, Preliminary Preparation: Collaborative calibration of UAVs, collection of basic information on the wind turbine blades to be tested, and planning of UAV flight paths;

[0013] Step S2, Background Acquisition: Without turning on the terahertz light source, the drone flies along the planned path and collects background image data of the wind turbine blades;

[0014] Step S3, Terahertz Transmission Imaging Acquisition: The terahertz light source is turned on, and the UAVs fly together along a preset path. The UAV equipped with the terahertz light source emits terahertz light towards the wind turbine blades. After the terahertz light passes through the wind turbine blades, the terahertz camera on another UAV receives the transmitted light signal and converts it into imaging data. During the imaging acquisition process, differential scanning is performed and data stitching is executed to generate an image. At the same time, temperature influence control and attitude influence control are performed during the image generation stage. After completion, the generated image is obtained, and the process proceeds to the next step.

[0015] Step S4, Data Preprocessing: Background Removal, Noise Reduction, and Extraction of Original Response Values;

[0016] Step S5, suspected defect area marking: Based on the extracted original response values ​​of the pixels, set a threshold range for the response values ​​of normal areas, and mark the areas where the pixels exceed the threshold range as suspected defect areas.

[0017] Step S6, Defect Detection and Identification: Input the marked suspected defect area data into the pre-trained defect identification model to accurately identify the suspected defect areas and detect the defect trend of the detected blades, and reflect the results simultaneously.

[0018] Step S7: Output the detection results.

[0019] Furthermore, the process of differential scanning and data stitching is as follows:

[0020] The blades under inspection are classified by region, and their structural, thickness, and importance characteristics are collected. The structural, thickness, and importance characteristics are then quantified and analyzed. The corresponding quantified parameters are compared with actual operating scenarios, i.e., the structural load-bearing, auxiliary load-bearing, and non-load-bearing methods are analyzed; the uniformity of structural thickness; and the blade maintenance probability when the quantified parameters of importance characteristics are abnormal.

[0021] Based on comparison, the blade area is divided into key areas, secondary areas, and routine areas; the scanning parameters of the terahertz detection system are dynamically adjusted according to the blade area type, including scanning step size, scanning speed, and signal acquisition frequency; through real-time area identification, parameters are automatically switched during the scanning process to avoid manual intervention.

[0022] Furthermore, seamless stitching is performed based on the segmented data of each type of blade region:

[0023] In the overlapping area of ​​adjacent scanning segments, the inherent structural features of the blade (such as the edge of the main beam and the abrupt change in thickness) are extracted as anchor points; the spatial coordinates of the segmented data are corrected based on the terahertz signal amplitude / phase characteristics of the anchor points; the terahertz amplitude of different scanning segments is normalized and compensated based on the signal mean of the overlapping area; the calibrated segmented data are spliced ​​according to coordinates to generate a complete defect map containing all regions.

[0024] Furthermore, the temperature effect control process is as follows:

[0025] When the UAV performs terahertz imaging, it collects real-time ambient temperature data using a temperature sensor and records temperature values ​​at various times. It then extracts the amplitude A of the terahertz time-domain signal main pulse at the same spatial relative position and time point between the tested blade and historical blades of the same type. 实时 (t) and A 对比 (t) is used to collect data; a reference signal drift model is constructed with temperature T and time t as variables;

[0026] A 实时 (t) refers to the peak voltage / current amplitude of the pulse signal under test at time t:

[0027] A 对比 (t) means: the reference master pulse amplitude refers to the peak voltage / current amplitude of the preset reference pulse signal with standard characteristics at time t.

[0028] Furthermore, the relative change is calculated to quantify the degree of deviation of the measured main pulse amplitude from the reference main pulse amplitude.

[0029] After obtaining the relative change, a threshold comparison is performed. If it is within the set threshold range, the pulse signal under test is determined to be valid; if it is not within the set range, the signal is determined to be interfered with or the equipment is abnormal, and the correction mechanism is triggered.

[0030] Furthermore, the attitude influence control method is as follows;

[0031] The system collects the vertical distance from the terahertz output probe of the UAV to the blade surface in real time; it also collects the attitude angle of the output probe and calculates the actual incident angle of the terahertz wave by combining it with the preset three-dimensional model of the blade surface; and it aligns the distance, incident angle and terahertz signal acquisition timestamp to ensure that the pose parameters and signal timing are matched.

[0032] Based on historical testing process logs, a calibration mapping relationship between distance and focusing lens displacement is pre-established. A miniature electric displacement platform is integrated on the output probe, which calls the mapping library based on the real-time distance to drive the focusing lens to adjust the displacement in real time. The average amplitude of the terahertz signal is monitored in real time. If the average amplitude is lower than the average amplitude threshold, the focusing parameters are dynamically corrected to form a closed-loop adjustment; otherwise, monitoring continues.

[0033] Furthermore, to address distance attenuation and angular phase distortion, a two-dimensional compensation is performed on the terahertz signal:

[0034] Based on the propagation attenuation model of terahertz waves, calculate the amplitude compensation coefficient corresponding to the distance;

[0035] The amplitude of the real-time detected main pulse is compensated and adjusted by the amplitude compensation coefficient; a fitting model of the incident angle-phase distortion compensation amount is established, and the phase of the terahertz signal is corrected in real time by the digital phase modulator to eliminate the phase distortion caused by the incident angle; the stability index of the compensated signal is calculated; if it does not meet the set normal range, the compensation process is repeated.

[0036] Furthermore, pixels exceeding the threshold range within the suspected defect area are identified and marked as abnormal points, while pixels within the threshold range are marked as normal points. Based on the normal points, the defect-free area is obtained, and the suspected defect area is compared with the defect-free area. The structural features, thickness features, and importance features of the corresponding area are collected, and the quantization parameters of each type of feature are collected simultaneously.

[0037] Quantitative parameters of the structural features corresponding to suspected defective areas and defect-free areas are determined, and the deviation of the structural feature parameters of the corresponding areas is determined based on parameter comparison.

[0038] If the deviation of structural characteristic parameters exceeds the structural characteristic parameter deviation threshold, the corresponding suspected defect area is set as a structural qualitative change trend; if the deviation of structural characteristic parameters does not exceed the structural characteristic parameter deviation threshold, the corresponding suspected defect area is set as a structural quantitative change trend.

[0039] The corresponding area location is then determined, and maintenance priorities are assigned based on the type of support method, with the priority order from high to low as: key areas, secondary areas, and routine areas.

[0040] Furthermore, the quantization parameters of the thickness features corresponding to the suspected defect area and the defect-free area are determined, and the deviation of the quantization parameters of the thickness features is obtained.

[0041] When the deviation of the quantization parameter of the thickness feature exceeds the thickness feature parameter deviation threshold, i.e. the time interval between adjacent detections, if the detection interval exceeds the time interval threshold, it is inferred that a slow high-span defect has appeared in the suspected defect area, the slow high-span defect type is set and used as the detection result; if the detection interval does not exceed the time interval threshold, it is inferred that a rapid high-span defect has appeared in the suspected defect area, the rapid high-span defect type is set and used as the detection result.

[0042] If the deviation of the quantization parameter of the thickness feature does not exceed the thickness feature parameter deviation threshold, i.e. the time interval between adjacent detections, if the detection interval exceeds the time interval threshold, it is inferred that a slow low-span defect has appeared in the suspected defect area, the slow low-span defect type is set and used as the detection result; if the detection interval does not exceed the time interval threshold, it is inferred that a rapid low-span defect has appeared in the suspected defect area, the rapid low-span defect type is set and used as the detection result.

[0043] Based on the type of the detection results, priorities are divided, and the order of priority from high to low is: rapid high-span defects, rapid low-span defects, slow high-span defects, and slow low-span defects.

[0044] Furthermore, the importance features corresponding to suspected defect areas and non-defect areas are determined, and the quantification parameters of the corresponding importance features are compared. The data deviation of the corresponding type of quantification parameter is recorded, and the quantification parameter type of the deviation of the importance feature and the corresponding data deviation are bound to the corresponding suspected defect area as the detection result.

[0045] Compared with the prior art, the beneficial effects of the present invention are:

[0046] 1. Based on the blade structure, thickness, and importance characteristics, regional classification is performed to achieve dynamic adjustment of scanning parameters. High-resolution, slow-speed scanning is used in key areas to ensure detection accuracy, while high-efficiency scanning is used in regular areas to improve detection efficiency, balancing accuracy and efficiency. Feature anchor point extraction, coordinate calibration, and signal compensation are combined to achieve seamless splicing of segmented data, avoiding defect omissions caused by splicing gaps, and generating a complete and accurate defect map.

[0047] 2. By collecting ambient temperature data in real time using a temperature sensor, a signal drift model based on temperature and time is constructed. The validity of the signal is determined by combining relative changes, enabling accurate identification of signal anomalies caused by temperature interference. An amplitude drift factor incorporates multiple environmental parameters and equipment operating time to normalize and correct abnormal signals, effectively compensating for amplitude deviations caused by temperature, humidity, and equipment aging, significantly improving imaging data stability and detection accuracy. A laser rangefinder and inertial measurement unit collaboratively collect distance and attitude angle data, and the actual incident angle is calculated using a three-dimensional blade model, achieving temporal matching between pose parameters and signals. Real-time adjustment of the focusing lens, distance-amplitude compensation, and incident angle-phase compensation form a closed-loop control, effectively eliminating the influence of distance attenuation and angle distortion on the signal, ensuring consistent imaging quality under different attitudes, and improving the anti-interference capability of the detection.

[0048] 3. By comparing the characteristic parameters of suspected defective areas and non-defective areas, the system can accurately determine the structural and thickness trends of defects, distinguishing between defect types with different development speeds and spans. By combining regional importance with defect trends to prioritize maintenance, and binding important characteristic deviation parameters, the system can provide a quantitative basis for maintenance decisions, predict defect development risks in advance, guide targeted maintenance, and prevent defects from expanding and causing safety accidents. Attached Figure Description

[0049] To facilitate understanding by those skilled in the art, the present invention will be further described below with reference to the accompanying drawings.

[0050] Figure 1 This is a system principle block diagram of the present invention;

[0051] Figure 2 This is a schematic diagram of terahertz imaging of wind turbine blades in a real-world scenario, as described in this invention. Figure 1 ;

[0052] Figure 3 This is a schematic diagram of terahertz imaging of wind turbine blades in a real-world scenario, as described in this invention. Figure 2 ;

[0053] Figure 4 This is a schematic diagram of a matrix scatter plot in a real-world scenario of the present invention. Figure 1 ;

[0054] Figure 5 This is a schematic diagram of a matrix scatter plot in a real-world scenario of the present invention. Figure 2 . Detailed Implementation

[0055] To enable those skilled in the art to better understand the present invention, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0056] In this document, the term "embodiment" means that a particular feature, structure, or characteristic described in connection with an embodiment may be included in at least one embodiment of the invention. The appearance of this phrase in various places throughout the specification does not necessarily refer to the same embodiment, nor is it a separate or alternative embodiment mutually exclusive with other embodiments. It will be explicitly and implicitly understood by those skilled in the art that the embodiments described herein can be combined with other embodiments.

[0057] This application utilizes terahertz transmission imaging technology for defect identification and detection of wind turbine blades. In practical scenarios, the following hardware is required: a dual-UAV system: employing two UAVs with stable hovering, precise positioning, and collaborative control capabilities. One UAV carries a terahertz light source, and the other carries a terahertz camera (detector). Terahertz light source: parameters are as follows: output frequency 110~365GHz, output power 18~23mW, polarization direction horizontal / vertical (adjustable), power supply voltage 220V, dimensions 167×112×109mm, continuous operation mode, capable of stable operation 7*24h. Terahertz camera (detector): employing a Terasense 64*64 resolution detector, possessing high sensitivity and fast data acquisition capabilities, accurately receiving terahertz light signals transmitted through the blades and converting them into electrical signals.

[0058] Please see Figure 1 As shown, an automatic defect region identification method applied to terahertz transmission imaging is described. The specific steps of the automatic defect region identification method are as follows:

[0059] Step S1: Preliminary Preparations

[0060] Two drones are calibrated collaboratively to ensure they maintain a preset distance and stable relative position during flight, preventing positional deviations from affecting imaging results. Simultaneously, basic information about the wind turbine blades under inspection (such as size, model, and installation location) is collected, and the drone flight path is planned to ensure coverage of the entire blade surface and key internal inspection areas. Collaborative calibration of the two drones ensures stable relative positions during flight, preventing misalignment between the light source and camera due to positional deviations, thus guaranteeing imaging accuracy. Collecting basic blade information and planning flight paths enables comprehensive coverage of the entire blade surface and key internal areas, eliminating missed inspections and laying the foundation for subsequent inspection work.

[0061] Step S2: Background Acquisition

[0062] Without turning on the terahertz light source, the drone equipped with a terahertz camera is controlled to fly along a planned path to collect background image data of the wind turbine blades. This data is used for subsequent background removal processing to eliminate the interference of ambient light and other noise on the detection results. Collecting background images without a terahertz light source provides data support for subsequent background removal processing, which can effectively eliminate the interference of ambient light, ground reflected light and other noise on the detection results, improve the signal-to-noise ratio of the imaging data, and create conditions for accurate extraction of defect signals.

[0063] Step S3: Terahertz transmission imaging acquisition

[0064] The terahertz light source is activated, and two drones are controlled to fly collaboratively along a preset path. One drone, equipped with a terahertz light source, emits terahertz light towards the wind turbine blades. After the terahertz light passes through the blades, it is received by a terahertz camera on the other drone and converted into imaging data. During flight, the drones maintain relative stability through precise positioning technology, ensuring the alignment accuracy between the light source and the camera. During image acquisition, differential scanning and data stitching are performed to generate an image. Temperature and attitude influence control are implemented during the image generation stage. Once completed, the generated image is obtained, and the process proceeds to the next step. The collaborative operation of the two drones achieves synchronized terahertz light emission and reception. Precise positioning technology ensures relative stability, guarantees the alignment accuracy between the light source and the camera, and improves the stability of transmitted signal acquisition. The combination of differential scanning and data stitching with temperature and attitude influence control generates comprehensive and high-quality imaging data, avoiding imaging distortion caused by environmental factors and changes in equipment attitude.

[0065] Step S4: Data Preprocessing

[0066] (1) Background removal processing: The difference between the acquired terahertz transmission imaging data and the previously acquired background image data is calculated to remove background noise interference;

[0067] (2) Noise reduction: The background-removed imaging data is denoised using filtering algorithms (such as median filtering and Gaussian filtering) to further optimize image quality and improve signal recognition.

[0068] (3) Extraction of raw response value: Extract the raw photosensitive response value of each pixel from the preprocessed imaging data. This value directly reflects the signal intensity of terahertz light after it is transmitted through the blade and is related to whether there are defects in the blade and the size and depth of the defects.

[0069] Step S5: Mark suspected defect areas, such as... Figure 2 , Figure 3 , Figure 4, Figure 5 As shown; where Figure 2 and Figure 3 These are schematic diagrams of terahertz imaging of wind turbine blades under different response values ​​and corresponding scenarios.

[0070] Based on the extracted original response values ​​of the pixels, a threshold range for the response values ​​of the normal area is set (this threshold range is determined based on a large amount of normal leaf data). The areas where the pixels exceed the threshold range are marked as suspected defect areas. At the same time, considering the influence of defect size and depth on the response values, cluster analysis is performed on the response values ​​of the pixels in the suspected defect areas to preliminarily distinguish different types of suspected defects.

[0071] Step S6: Defect Detection and Identification: Input the data of the marked suspected defect areas (elliptical and rectangular areas) into the pre-trained defect identification model. This model is trained based on a large amount of wind turbine transmission imaging data of different defect types (such as surface cracks, internal delamination, material aging, etc.), different sizes, and different depths. It can accurately identify suspected defect areas, determine the specific type, size, and depth of the defect, and output the final defect detection result. It also performs defect trend detection on the detected blades and reflects it in the detection results.

[0072] Step S7: Output the detection results.

[0073] Example 2

[0074] In the previous embodiment, differential scanning and data stitching are further disclosed to improve the accuracy of terahertz transmission imaging;

[0075] The tested blades were classified by region, and their structural, thickness, and importance characteristics were collected. The structural characteristics were quantified by parameters such as total blade length (m): the length from the blade root to the blade tip; and the blade swept area. The thickness characteristics were quantified by parameters such as overall thickness distribution (mm): the thickness values ​​at the blade root, blade body, and blade tip. The importance characteristics were quantified by parameters such as ultimate load (kN): the maximum static load the blade could withstand; and fatigue life (cycles): the number of cycles it could withstand under the design load (usually requiring 10 cycles). 7 (more than once);

[0076] The structural features, thickness features, and importance features are quantitatively analyzed, and the corresponding quantitative parameters are compared with the actual operating scenarios. This includes analyzing the structural load-bearing, auxiliary load-bearing, and non-load-bearing methods; structural thickness uniformity; and the blade maintenance probability when the quantitative parameters of importance features are abnormal.

[0077] Based on comparison, the blade area is divided into key areas, secondary areas, and regular areas; key areas include the main beam and leading edge (structural load-bearing / easily damaged); secondary areas include the trailing edge (auxiliary load-bearing); and regular areas include the web (non-load-bearing / uniform thickness). In the application scenario of this application, the regional classification results of the target blade can be automatically loaded before detection through a preset regional database (storing the regional division coordinates of different blade models) as the basis for configuring scanning parameters.

[0078] The scanning parameters of the terahertz detection system are dynamically adjusted based on the blade region type. These parameters include the scanning step size, scanning speed, and signal acquisition frequency. Through real-time region identification, the parameters are automatically switched during the scanning process, avoiding manual intervention.

[0079] In real-world usage scenarios,

[0080] Key area step size = 0.5mm (high resolution); Secondary area step size = 1mm (medium resolution); Regular area step size = 2mm (regular resolution).

[0081] Speed ​​in key areas = 0.1 m / s (slow scan, high resolution); speed in secondary areas = 0.3 m / s (medium speed); speed in regular areas = 0.8 m / s (fast scan, high efficiency).

[0082] Match the scanning speed to ensure that the sampling point density in each area meets the resolution requirements;

[0083] Seamlessly stitch together the segmented data of each type of blade region:

[0084] 1. Feature anchor point extraction: In the overlapping area (width ≥ 5cm) of adjacent scan segments, extract the inherent structural features of the blade (such as the edge of the main beam and the abrupt change in thickness) as anchor points;

[0085] 2. Coordinate fine calibration: Based on the amplitude / phase characteristics of the terahertz signal at the anchor point, the spatial coordinates of the segmented data are corrected for errors (correction accuracy ≤ 0.1mm).

[0086] 3. Signal consistency compensation: Based on the signal mean of the overlapping area, the terahertz amplitude of different scanning segments is normalized and compensated (the signal deviation after compensation is ≤2%).

[0087] 4. Global map generation: The calibrated segmented data is stitched together according to coordinates to generate a complete defect map containing all regions (supporting quantitative labeling of defect location and size).

[0088] Example 3

[0089] Based on the above embodiments, and referring to the traditional method of historical detection process, the temperature influence control is improved to effectively reduce the impact of temperature fluctuations during the image acquisition stage.

[0090] When the UAV performs terahertz imaging, it collects real-time ambient temperature data using a temperature sensor and records temperature values ​​at various times. It then extracts the amplitude A of the terahertz time-domain signal main pulse at the same spatial relative position and time point between the tested blade and historical blades of the same type. 实时 (t) and A 对比 (t). Construct a reference signal drift model ΔA(T,t) with temperature T and time t as variables;

[0091] Specifically: the amplitude A of the main pulse to be measured 实时 (t) refers to the peak voltage / current amplitude of the pulse signal under test at time t, and is a key physical quantity characterizing the intensity, energy, or effective range of the pulse signal under test. Its mathematical expression can be defined as:

[0092] ;

[0093] U 实时 (τ) represents the instantaneous voltage / current value of the pulse signal under test at time τ;

[0094] [t0,t1] is the effective duration window of a single main pulse;

[0095] Reference main pulse amplitude A 对比 (t) refers to the peak voltage / current amplitude of a preset reference pulse signal with standard characteristics at time t, which serves as a comparison benchmark for the amplitude of the pulse to be tested and is used to determine whether the signal to be tested meets the technical requirements.

[0096] Its sources include two types:

[0097] Fixed reference amplitude: a constant threshold preset by the system (such as A). 对比 (t)=A0, where A0 is a constant), suitable for static calibration scenarios;

[0098] Dynamic reference amplitude: An amplitude generated in real time by a reference device or adaptively adjusted according to operating conditions (e.g., A). 对比 (t)=f(T,P), where T is temperature and P is pressure), suitable for complex dynamic scenarios.

[0099] The relative change R(x, y) is calculated to quantize the amplitude A of the main pulse under test. 实时 (x,y) relative to the reference master pulse amplitude A 对比 The degree of deviation of (t) is given by the following formula:

[0100] ;

[0101] (x,y): Spatial / operating condition characteristic parameters of the pulse signal to be measured (such as the point coordinates of the blade being tested);

[0102] A 实时 (x,y) represents the instantaneous amplitude of the main pulse under test, acquired under characteristic parameters (x,y), with dimensions equal to A. 对比 (t) consistent;

[0103] A 对比 (t) The reference main pulse amplitude at time t can be a fixed threshold (static calibration scenario) or a dynamic real-time value (operating condition adaptive scenario).

[0104] R(x, y) is a dimensionless relative change. A positive value indicates that the measured amplitude is higher than the reference value, a negative value indicates that it is lower than the reference value, and a zero value indicates a perfect match.

[0105] After obtaining R(x, y), a threshold comparison is performed. If it is within the set threshold range, the pulse signal under test is determined to be valid; if it is not within the set range, the signal is determined to be interfered with or the equipment is abnormal, and the correction mechanism is triggered.

[0106] When R(x,y) exceeds the threshold range, A needs to be adjusted using a drift model. 实时 (x,y) is normalized to compensate for amplitude drift caused by environmental interference (temperature, humidity), equipment aging, or transmission loss.

[0107] Define an amplitude drift factor K(x,y,t), which is related to environmental parameters T(t) (temperature), H(t) (humidity), and equipment operating time tr, and its expression is:

[0108] K(x,y,t)=k0+k1×T(t)+k2×H(t)+k3×ln(tr+1)

[0109] Where: k0, k1, k2, k3: fitting coefficients calibrated through a large number of experiments (which can be solved by the least squares method);

[0110] k0: Base drift coefficient, characterizing the inherent drift when there is no environmental disturbance and equipment aging;

[0111] k1, k2: Environmental sensitivity coefficients, which characterize the weights of temperature and humidity on the amplitude;

[0112] k3: Aging attenuation coefficient, characterizing the attenuation effect of equipment operating time on amplitude.

[0113] After obtaining the amplitude drift factor, the amplitude of the main pulse under test that is abnormal is divided and the relative change is calculated repeatedly. If the relative change is within the set threshold range, the amplitude of the main pulse under test is adjusted to the normal state.

[0114] Example 4

[0115] In the above embodiments, the pose influence control is improved during the image generation stage to further enhance the accuracy of image acquisition; the pose influence control method is as follows:

[0116] The vertical distance from the UAV's terahertz output probe to the blade surface is collected in real time using a laser rangefinder; the attitude angles (pitch angle and roll angle) of the output probe are collected using an inertial measurement unit (IMU), and the actual incident angle of the terahertz wave is calculated by combining the preset three-dimensional model of the blade surface; the distance, incident angle and terahertz signal acquisition timestamp are aligned to ensure that the pose parameters and signal timing are matched.

[0117] Based on historical testing process logs, a calibration mapping relationship between distance and focusing lens displacement is pre-established (the optimal focusing position at different distances is obtained through historical testing); a miniature electric displacement platform is integrated on the output probe, which calls the mapping library according to the real-time distance to drive the focusing lens to adjust the displacement in real time;

[0118] The average amplitude of the terahertz signal is monitored in real time. If the average amplitude is lower than the average amplitude threshold, the focusing parameters are dynamically adjusted to form a closed-loop adjustment; otherwise, monitoring continues.

[0119] To address distance attenuation and angular phase distortion, a two-dimensional compensation method is used for terahertz signals:

[0120] Based on the propagation attenuation model of terahertz waves, calculate the amplitude compensation coefficient corresponding to the distance;

[0121] ;

[0122] d(t) is the vertical distance at time t; α is the attenuation coefficient of the terahertz wave in the air medium, which is determined experimentally; e is the natural constant.

[0123] Through the amplitude compensation coefficient K d (t) The amplitude of the real-time detection main pulse is compensated and adjusted, specifically by multiplication to adjust the coefficients;

[0124] A fitting model for the incident angle-phase distortion compensation amount is established (based on electromagnetic simulation to obtain the phase change law under different incident angles). The phase of the terahertz signal is corrected in real time by a digital phase modulator to eliminate the phase distortion caused by the incident angle.

[0125] Calculate the stability indicators of the compensated signal, including amplitude fluctuation coefficient and phase deviation; if the normal range is not met, repeat the compensation process to ensure signal stability.

[0126] Example 5

[0127] In Example 1, based on defect detection and identification, defect trend detection is performed on the detected blades, and the results are reflected simultaneously to improve the pertinence and decision-making of blade maintenance.

[0128] Pixels that exceed the threshold range within the suspected defect area are identified and marked as abnormal points, while pixels that do not exceed the threshold range are marked as normal points.

[0129] Defect-free areas are obtained from normal locations, and suspected defective areas are compared with defect-free areas. Structural features, thickness features, and importance features of the corresponding areas are collected, and quantitative parameters of each type of feature are collected simultaneously.

[0130] Quantitative parameters of the structural features corresponding to suspected defective areas and defect-free areas are determined. Based on parameter comparison, the deviation of structural feature parameters of the corresponding areas is determined, specifically: blade length deviation or swept area deviation.

[0131] If the deviation of structural characteristic parameters exceeds the structural characteristic parameter deviation threshold, the corresponding suspected defect area is set as a structural qualitative change trend; if the deviation of structural characteristic parameters does not exceed the structural characteristic parameter deviation threshold, the corresponding suspected defect area is set as a structural quantitative change trend.

[0132] And determine the carrying method of the corresponding area, and divide the maintenance priority according to the carrying method type, that is, the priority order from high to low is: key area, secondary area, and regular area;

[0133] Determine the quantization parameters of the thickness features corresponding to the suspected defect area and the defect-free area, and obtain the deviation of the quantization parameters of the thickness features;

[0134] When the deviation of the quantization parameter of the thickness feature exceeds the thickness feature parameter deviation threshold, i.e. the time interval between adjacent detections, if the detection interval exceeds the time interval threshold, it is inferred that a slow high-span defect has appeared in the suspected defect area, the slow high-span defect type is set and used as the detection result; if the detection interval does not exceed the time interval threshold, it is inferred that a rapid high-span defect has appeared in the suspected defect area, the rapid high-span defect type is set and used as the detection result.

[0135] If the deviation of the quantization parameter of the thickness feature does not exceed the thickness feature parameter deviation threshold, i.e. the time interval between adjacent detections, if the detection interval exceeds the time interval threshold, it is inferred that a slow low-span defect has appeared in the suspected defect area, the slow low-span defect type is set and used as the detection result; if the detection interval does not exceed the time interval threshold, it is inferred that a rapid low-span defect has appeared in the suspected defect area, the rapid low-span defect type is set and used as the detection result.

[0136] Based on the type of the detection results, priorities are divided, and the order of priority from high to low is: rapid high-span defects, rapid low-span defects, slow high-span defects, and slow low-span defects.

[0137] The importance features corresponding to suspected defect areas and non-defect areas are determined, and the quantitative parameters of the corresponding importance features are compared. The data deviation of the corresponding type of quantitative parameter is recorded. The quantitative parameter type of the deviation of the importance feature and the corresponding data deviation are bound to the corresponding suspected defect area as the detection result. When performing maintenance on suspected defect areas, this is used as a priority evaluation index and also as a standard parameter for whether to shut down the system.

[0138] Thresholds, preset values, preset ranges, etc. are set for result comparison and analysis to determine whether they are good or bad. The value of these thresholds is determined by a combination of large-scale model analysis of sample data and human experience. They can also be adjusted appropriately based on seasonal or common-sense influences.

[0139] The preferred embodiments of the present invention disclosed above are merely illustrative of the invention. These preferred embodiments do not exhaustively describe all details, nor do they limit the invention to any specific implementation. Clearly, many modifications and variations can be made based on the content of this specification. This specification selects and specifically describes these embodiments to better explain the principles and practical applications of the invention, thereby enabling those skilled in the art to better understand and utilize the invention. The invention is limited only by the claims and their full scope and equivalents.

Claims

1. An automatic defect region identification method applied to terahertz transmission imaging, characterized in that, The steps of the automatic defect area identification method are as follows: Step S1, Preliminary Preparation: Collaborative calibration of UAVs, collection of information on the wind turbine blades to be tested, and planning of UAV flight paths; Step S2, Background Acquisition: Without turning on the terahertz light source, the drone flies along the planned path and collects background image data of the wind turbine blades; Step S3, Terahertz Transmission Imaging Acquisition: The terahertz light source is turned on, and the UAVs fly together along a preset path. The UAV equipped with the terahertz light source emits terahertz light towards the wind turbine blades. After the terahertz light passes through the wind turbine blades, the terahertz camera on another UAV receives the transmitted light signal and converts it into imaging data. During the imaging acquisition process, differential scanning is performed and data stitching is executed to generate an image. At the same time, temperature influence control and attitude influence control are performed during the image generation stage. After completion, the generated image is obtained, and the process proceeds to the next step. Step S4, Data Preprocessing: Background Removal, Noise Reduction, and Extraction of Original Response Values; Step S5, suspected defect area marking: Based on the extracted original response values ​​of the pixels, set a threshold range for the response values ​​of the normal area, and mark the areas where the pixels exceed the threshold range as suspected defect areas. Step S6, Defect Detection and Identification: Input the marked suspected defect area data into the pre-trained defect identification model to accurately identify the suspected defect areas and detect the defect trend of the detected blades, and reflect the results simultaneously. Step S7: Output the detection results.

2. The method for automatic defect region identification applied to terahertz transmission imaging according to claim 1, characterized in that, The process of differential scanning and data stitching is as follows: The tested blades were classified by region, and their structural, thickness, and importance characteristics were collected. The structural features, thickness features, and importance features are analyzed using quantitative parameters. The corresponding quantitative parameters are compared with actual operating scenarios, i.e., the structural load-bearing, auxiliary load-bearing, and non-load-bearing methods are analyzed; and the uniformity of structural thickness is analyzed. Importance features are represented as the probability of blade maintenance during anomalies; Based on comparison, the blade area is divided into key areas, secondary areas, and routine areas; the scanning parameters of the terahertz detection system are dynamically adjusted according to the blade area type, including scanning step size, scanning speed, and signal acquisition frequency; through real-time area identification, parameters are automatically switched during the scanning process to avoid manual intervention.

3. The automatic defect region identification method for terahertz transmission imaging according to claim 2, characterized in that, Seamlessly stitch together the segmented data of each type of blade region: Extract the inherent structural features of the blade as anchor points in the overlapping area of ​​adjacent scan segments; Based on the amplitude / phase characteristics of the terahertz signal at the anchor point, the spatial coordinates of the segmented data are corrected for errors. Based on the signal mean of the overlapping area, the terahertz amplitude of different scanning segments is normalized and compensated. The calibrated segmented data are then stitched together according to the coordinates to generate a complete defect map containing all regions.

4. The method for automatic defect region identification applied to terahertz transmission imaging according to claim 1, characterized in that, The temperature effect control process is as follows: When the UAV performs terahertz imaging, it collects real-time ambient temperature data using a temperature sensor and records temperature values ​​at various times. It then extracts the amplitude A of the terahertz time-domain signal main pulse at the same spatial relative position and time point between the tested blade and historical blades of the same type. 实时 (t) and A 对比 (t) is used to collect data; a reference signal drift model is constructed with temperature T and time t as variables; A 实时 (t) refers to the peak voltage / current amplitude of the pulse signal under test at time t: A 对比 (t) means: the reference master pulse amplitude refers to the peak voltage / current amplitude of the preset reference pulse signal with standard characteristics at time t.

5. The automatic defect region identification method for terahertz transmission imaging according to claim 4, characterized in that, The relative change is calculated to quantify the degree of deviation of the measured master pulse amplitude from the reference master pulse amplitude. After obtaining the relative change, a threshold comparison is performed. If it is within the set threshold range, the pulse signal under test is determined to be valid; if it is not within the set range, the signal is determined to be interfered with or the equipment is abnormal, and the correction mechanism is triggered.

6. The automatic defect region identification method for terahertz transmission imaging according to claim 5, characterized in that, The attitude influence control methods are as follows; The system collects the vertical distance from the terahertz output probe of the UAV to the blade surface in real time; it also collects the attitude angle of the output probe and calculates the actual incident angle of the terahertz wave by combining it with the preset three-dimensional model of the blade surface; and it aligns the distance, incident angle and terahertz signal acquisition timestamp to ensure that the pose parameters and signal timing are matched. Based on historical testing process logs, a calibration mapping relationship between distance and focusing lens displacement is pre-established. A miniature electric displacement platform is integrated on the output probe, which calls the mapping library based on the real-time distance to drive the focusing lens to adjust the displacement in real time. The average amplitude of the terahertz signal is monitored in real time. If the average amplitude is lower than the average amplitude threshold, the focusing parameters are dynamically corrected to form a closed-loop adjustment; otherwise, monitoring continues.

7. The method for automatic defect region identification applied to terahertz transmission imaging according to claim 6, characterized in that, To address distance attenuation and angular phase distortion, a two-dimensional compensation method is used for terahertz signals: Based on the propagation attenuation model of terahertz waves, calculate the amplitude compensation coefficient corresponding to the distance; The amplitude of the real-time detected main pulse is compensated and adjusted by the amplitude compensation coefficient; a fitting model of the incident angle-phase distortion compensation amount is established, and the phase of the terahertz signal is corrected in real time by the digital phase modulator to eliminate the phase distortion caused by the incident angle; the stability index of the compensated signal is calculated; if it does not meet the set normal range, the compensation process is repeated.

8. The method for automatic defect region identification applied to terahertz transmission imaging according to claim 1, characterized in that, Pixels exceeding the threshold range within the suspected defect area are identified and marked as abnormal points, while pixels within the threshold range are marked as normal points. Defect-free areas are obtained based on the normal points, and the suspected defect area is compared with the defect-free area. Structural features, thickness features, and importance features of the corresponding area are collected, and quantization parameters of each type of feature are collected simultaneously. Quantitative parameters of the structural features corresponding to suspected defective areas and defect-free areas are determined, and the deviation of the structural feature parameters of the corresponding areas is determined based on parameter comparison. If the deviation of structural characteristic parameters exceeds the structural characteristic parameter deviation threshold, the corresponding suspected defect area will be set as a structural qualitative change trend. If the deviation of the structural characteristic parameters does not exceed the structural characteristic parameter deviation threshold, the corresponding suspected defect area will be set as the structural quantitative change trend. The corresponding area location is then determined, and maintenance priorities are assigned based on the type of support method, with the priority order from high to low as: key areas, secondary areas, and routine areas.

9. The method for automatic defect region identification applied to terahertz transmission imaging according to claim 8, characterized in that, Determine the quantization parameters of the thickness features corresponding to the suspected defect area and the defect-free area, and obtain the deviation of the quantization parameters of the thickness features; When the deviation of the quantization parameter of the thickness feature exceeds the thickness feature parameter deviation threshold, i.e. the time interval between adjacent detections, if the detection interval exceeds the time interval threshold, it is inferred that a slow high-span defect has appeared in the suspected defect area, the slow high-span defect type is set and used as the detection result; if the detection interval does not exceed the time interval threshold, it is inferred that a rapid high-span defect has appeared in the suspected defect area, the rapid high-span defect type is set and used as the detection result. If the deviation of the quantization parameter of the thickness feature does not exceed the thickness feature parameter deviation threshold, i.e. the time interval between adjacent detections, if the detection interval exceeds the time interval threshold, it is inferred that a slow low-span defect has appeared in the suspected defect area, the slow low-span defect type is set and used as the detection result; if the detection interval does not exceed the time interval threshold, it is inferred that a rapid low-span defect has appeared in the suspected defect area, the rapid low-span defect type is set and used as the detection result. Based on the type of the detection results, priorities are divided, and the order of priority from high to low is: rapid high-span defects, rapid low-span defects, slow high-span defects, and slow low-span defects.

10. The method for automatic defect region identification applied to terahertz transmission imaging according to claim 9, characterized in that, The importance features corresponding to suspected defect areas and non-defect areas are determined, and the quantification parameters of the corresponding importance features are compared. The data deviation of the corresponding type of quantification parameter is recorded. The quantification parameter type of the deviation of the importance feature and the corresponding data deviation are bound to the corresponding suspected defect area as the detection result.

Images