Multi-dimensional sensing and AI combined fan blade defect detection system and method thereof

Abstract

The invention discloses a multi-dimensional sensing and AI combined fan blade defect detection system and method, and relates to the technical field of fan blade detection.The system comprises a wind driven generator body and further comprises a multi-dimensional sensing module, an edge calculation module and a cloud AI module, and the wind driven generator body comprises a tower base, a generator head, a flow guide cover and blades; the generator head is installed at the top of the tower base, the flow guide cover is installed at the front end of the generator head, the blade is fixed to the flow guide cover, a fixing bolt is arranged between the blade and the flow guide cover, the multi-dimensional sensing module comprises a voiceprint sensor, an ultrasonic bolt axial force detection sensor and a temperature and vibration integrated sensor, and the voiceprint sensor is installed in the blade and on the tower base. The system has the advantages of multi-dimensional data acquisition, real-time monitoring, automatic cleaning and intelligent diagnosis, and solves the problems of single sensing dimension, poor environmental adaptability, slow fault response and lack of bolt axial force monitoring in the prior art.

Description

Technical Field

[0001] This invention relates to the field of wind turbine blade inspection technology, specifically to a wind turbine blade defect detection system and method that combines multi-dimensional sensing and AI. Background Technology

[0002] Wind power, as an important component of clean energy, has been widely used globally in recent years. With the continuous increase in the capacity of individual wind turbine units and the increasing complexity of operating environments, the reliability and safety of key wind turbine components have become increasingly prominent. Among them, wind turbine blades, as the core component of energy conversion, operate for extended periods in harsh environments of high altitude, high speed, and variable loads, making them highly susceptible to various factors such as lightning strikes, dust erosion, moisture intrusion, fatigue damage, and icing, leading to structural defects such as cracks, delamination, and fractures. Serious failures can not only cause huge economic losses but may even trigger catastrophic accidents such as blade breakage and tower collapse, threatening the safety of personnel and equipment.

[0003] Currently, health monitoring of wind turbine blades mainly relies on manual inspections, periodic shutdown checks, and online monitoring methods based on single sensors. Manual inspections are costly, inefficient, and limited by weather and geographical conditions, making it difficult to achieve high-frequency, comprehensive real-time monitoring. Traditional online monitoring technologies often use vibration sensors or strain gauges for localized monitoring, offering only a single sensing dimension and failing to comprehensively reflect the overall health status of the blades, especially limiting their ability to identify early, minor defects (such as internal delamination and microcrack propagation). Furthermore, bolted connections, as crucial structural connection points between the blades and the hub, pose a significant risk of increased blade vibration or even blade detachment if the preload loosens or fails. However, existing technologies lack effective means for long-term, accurate, and non-destructive online monitoring of bolt axial forces. Therefore, a wind turbine blade defect detection system and method combining multi-dimensional sensing and AI are needed to address these issues. Summary of the Invention

[0004] The purpose of this invention is to provide a wind turbine blade defect detection system and method that combines multi-dimensional sensing and AI. It has the advantages of multi-dimensional data acquisition, real-time monitoring, automatic cleaning and intelligent diagnosis, and solves the problems of single sensing dimension, poor environmental adaptability, slow fault response and lack of bolt axial force monitoring in the prior art.

[0005] To achieve the above objectives, the present invention provides the following technical solution: a wind turbine blade defect detection system combining multidimensional sensing and AI, comprising a wind turbine main body, and further comprising a multidimensional sensing module, an edge computing module and a cloud AI module;

[0006] The wind turbine body includes a tower, a generator head, a fairing, and blades. The generator head is installed on the top of the tower, the fairing is installed at the front end of the generator head, and the blades are fixed to the fairing. Fixing bolts are provided between the blades and the fairing. The multi-dimensional sensing module includes an acoustic fingerprint sensor, an ultrasonic bolt axial force detection sensor, and a temperature and vibration integrated sensor. The acoustic fingerprint sensor is installed inside the blades and on the tower, the ultrasonic bolt axial force detection sensor is installed on the fixing bolts, and the temperature and vibration integrated sensor is installed inside the blades.

[0007] As a preferred embodiment of the multi-dimensional sensing and AI-integrated wind turbine blade defect detection system of the present invention, it further includes an air blowing dust removal mechanism. The air blowing dust removal mechanism includes a sealing cover, a winding drum, a drive motor, a high-pressure air pump, and an air blowing airbag. The sealing cover is fixedly installed inside the tower base. The acoustic sensor is installed at the front end of the sealing cover. The winding drum is installed inside the sealing cover and rotatably connected to it. The drive motor is installed inside the sealing cover to drive the winding drum. The air blowing airbag is installed on the winding drum. The high-pressure air pump is installed inside the sealing cover to supply air to the air blowing airbag.

[0008] As a preferred embodiment of the multi-dimensional sensing and AI-integrated wind turbine blade defect detection system of the present invention, the front end face of the sealing cover is provided with an equipment compartment, the winding drum, the drive motor and the high-pressure air pump are all installed in the equipment compartment, the side end face of the equipment compartment is provided with a guide groove, and the top of the equipment compartment is provided with a sealing cover that slides in cooperation with the guide groove.

[0009] In a preferred embodiment of the multi-dimensional sensing and AI-integrated wind turbine blade defect detection system of the present invention, the drive motor is fixedly installed inside the equipment compartment, a rack is fixedly installed on the lower end face of the sealing cover, and a gear is fixedly connected to the output shaft of the drive motor.

[0010] As a preferred embodiment of the multi-dimensional sensing and AI-integrated wind turbine blade defect detection system of the present invention, the equipment compartment is provided with a bearing seat, the bearing seat is provided with a first shaft hole for rotatably connecting with the winding drum, the output shaft of the drive motor is provided with a first synchronous pulley, the side end face of the winding drum is provided with a second synchronous pulley, and the first and second synchronous pulleys are provided with synchronous belts for transmission.

[0011] As a preferred embodiment of the multi-dimensional sensing and AI-integrated wind turbine blade defect detection system of the present invention, the high-pressure air pump is fixedly installed in the equipment compartment, the air outlet of the high-pressure air pump is provided with an air supply pipe, the end of the air supply pipe is provided with an air supply cover that cooperates with the winding drum, the winding drum is provided with a first air hole that communicates with the air supply cover, the top of the winding drum is provided with a second air hole that communicates with the blowing air bag, and the side end face of the air supply cover is provided with a second shaft hole that cooperates with the winding drum.

[0012] As a preferred embodiment of the multi-dimensional sensing and AI-integrated wind turbine blade defect detection system of the present invention, the blowing airbag has a circular structure, and the bottom of the blowing airbag is provided with an inflation hole that communicates with the winding drum. The acoustic sensor includes a protective cover and a speaker unit. The protective cover (2011) is provided with a through hole, and the blowing airbag is uniformly provided with blowing heads corresponding to the positions of the through holes.

[0013] As a preferred embodiment of the multi-dimensional sensing and AI-integrated wind turbine blade defect detection system of the present invention, the front end face of the sealing cover is provided with a rain shield, the lower end face of the rain shield is provided with a magnet, and the top of the air-blowing bag is provided with an iron piece corresponding to the position of the magnet.

[0014] As a preferred embodiment of the multi-dimensional sensing and AI-integrated wind turbine blade defect detection system of the present invention, the number of temperature and vibration integrated sensors inside the blade is three, and the three sets of temperature and vibration integrated sensors are respectively located at the top, middle and bottom of the blade.

[0015] A method for detecting defects in wind turbine blades by combining multidimensional sensing and AI includes the following steps:

[0016] Step 1: Multi-dimensional data acquisition, deploying acoustic fingerprint sensors, ultrasonic bolt axial force detection sensors, and temperature and vibration integrated sensors to acquire blade data in real time;

[0017] Step 2: Real-time preprocessing of the edge layer, adaptive spectral subtraction denoising of the acoustic signal, baseline calibration and extraction of time-domain features of the temperature and vibration signal, and temperature compensation correction of the bolt axial force signal.

[0018] Step 3: Preliminary edge layer diagnosis. The preprocessed data is analyzed using a lightweight AI model to identify sudden drops in bolt axial force and abnormal blade vibration and trigger alarms. Non-abnormal data is stored locally.

[0019] Step 4: Cloud-based deep diagnostics, integrating multi-source preprocessed data to generate health feature vectors, using a transfer learning model to identify defect types and locations, and combining a random forest model to predict remaining service life;

[0020] Step 5: Operation and maintenance decision-making and closed loop. Generate early warning information and push it to the operation and maintenance platform. Optimize the inspection plan and record defect development data for model iteration.

[0021] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0022] 1. This invention utilizes integrated temperature and vibration sensors deployed at key locations within the blade to capture vibration anomalies and temperature changes in real time. Different types of high-sensitivity acoustic signature sensors are deployed in the blade's sealed cavity and the tower base to monitor abnormal noises. Ultrasonic bolt axial force sensors are installed at key connection points to monitor the tightening status. This constructs a multi-dimensional, comprehensive sensing network for blade structural health, bolt connection status, and the operating environment. The collected raw data undergoes targeted preprocessing in an edge computing module deployed at the bottom of the tower, including acoustic signature denoising, temperature and vibration baseline calibration, bolt axial force temperature compensation, and real-time inference analysis using a lightweight AI model. This enables rapid identification of sudden and severe faults such as sudden drops in bolt axial force and abnormal peak vibrations in the blade, immediately triggering alarms and addressing key risks. With a response time of up to seconds, non-urgent data is preprocessed and uploaded to the cloud-based AI module. This module integrates all sensor information through a multi-source data fusion unit and utilizes deep diagnostic model units, such as transfer learning CNN to extract voiceprint features and random forest to assess severity, to perform more complex defect type identification, localization, and remaining life prediction. This layered intelligent processing architecture of "real-time edge warning + cloud-based deep diagnostics" effectively solves the problems of traditional detection methods, such as manual inspection and single-sensor monitoring, which suffer from slow response, high false negative rates, and difficulty in early detection of hidden defects and assessment of damage. It greatly improves the accuracy, timeliness, and comprehensiveness of wind turbine blade defect detection, providing a reliable basis for preventing catastrophic failures and developing precise maintenance plans.

[0023] 2. This invention integrates an air-blowing dust removal mechanism. Its core structure includes a sealed cover housing the acoustic sensor, a retractable air-blowing bladder, a motor driving the retraction, a high-pressure air pump providing the air source, and a linked opening and closing sealing cover. When cleaning is required, the drive motor, via a synchronous belt mechanism, rotates the winding drum to release the air-blowing bladder. Simultaneously, a gear on the motor's output shaft drives a rack on the sealing cover, causing the cover to slide open synchronously. Subsequently, the high-pressure air pump inflates the winding drum, and the airflow enters the annular air-blowing bladder through the drum's internal channels, causing it to inflate. At this time, the evenly distributed air-blowing heads on the bladder are precisely aligned with the through-holes of the acoustic sensor's protective cover, and the high-pressure air... The airflow is directly blown into the interior of the protective cover, efficiently removing accumulated sand and dust, realizing the automation and integration of dust removal. Its sealed equipment compartment design also effectively prevents dust and rainwater from entering the internal equipment. This automated dust removal function fundamentally solves the key pain point of the acoustic fingerprint sensor exposed at the front of the tower base in the harsh environment of the wind farm, which is prone to decreased sensitivity due to sand and dust accumulation and failure due to protective cover blockage. It ensures the long-term stability and reliability of the data in the important dimension of acoustic fingerprint perception, thereby maintaining the continuity and stability of the detection accuracy of the entire system and significantly reducing false alarms, missed alarms and frequent maintenance needs caused by sensor contamination. Attached Figure Description

[0024] Figure 1 This is a schematic diagram of the overall structure of the present invention;

[0025] Figure 2 This is a side sectional view of the present invention;

[0026] Figure 3 This is a schematic diagram of the air blowing dust removal mechanism of the present invention in the stopped state;

[0027] Figure 4 This is a schematic diagram of the air blowing dust removal mechanism in the start-up state of the present invention;

[0028] Figure 5 This is a side sectional view of the air blowing dust removal mechanism of the present invention in the start-up state;

[0029] Figure 6 This is a schematic diagram of the air blowing dust removal mechanism of the present invention;

[0030] Figure 7 This is a cross-sectional view of the blade of the present invention;

[0031] Figure 8 This is a schematic diagram of the fixing bolt mechanism of the present invention;

[0032] Figure 9 This is a cross-sectional view of the winding drum and the blowing airbag of the present invention in a mating state;

[0033] Figure 10 For the present invention Figure 1 Enlarged view of point A in the middle;

[0034] Figure 11 For the present invention Figure 5 Enlarged view at point B in the middle;

[0035] Figure 12 For the present invention Figure 9 Enlarged view at point C;

[0036] Figure 13 For the present invention Figure 5 Enlarged view at point D;

[0037] Figure 14 This is a flowchart of the detection method of the present invention;

[0038] Figure 15 This is a flowchart of the cloud-based AI module noise processing method of the present invention.

[0039] In the diagram: 1. Wind turbine main body; 101. Tower base; 102. Generator head; 103. Shielding; 104. Blades; 105. Fixing bolts; 2. Multi-dimensional sensing module; 201. Acoustic sensor; 2011. Protective cover; 2012. Speaker unit; 2013. Through hole; 202. Ultrasonic bolt axial force detection sensor; 203. Temperature and vibration integrated sensor; 3. Air blowing dust removal mechanism; 301. Sealing cover; 302. Equipment compartment; 303. Sealing cap; 304. Rain shield; 3041. Magnet; 305. Winding drum; 306. Bearing housing; 3161. First shaft hole; 307. Drive motor; 308. Guide groove; 309. Rack; 310. Gear; 311. First synchronous pulley; 312. Second synchronous pulley; 313. Synchronous belt; 314. Air supply cover; 3141. Second shaft hole; 315. Air supply pipe; 3151. First air hole; 3152. Second air hole; 316. Air bladder; 3161. Air inflator; 3162. Iron sheet; 3163. Inflation hole; 317. High-pressure air pump. Detailed Implementation

[0040] Example 1

[0041] Please see Figures 1-13 A wind turbine blade defect detection system combining multidimensional sensing and AI includes a wind turbine main body 1, a multidimensional sensing module 2, an edge computing module and a cloud AI module.

[0042] The wind turbine main body 1 includes a tower base 101, a generator head 102, a fairing 103, and blades 104. The generator head 102 is installed on the top of the tower base 101, the fairing 103 is installed at the front end of the generator head 102, and the blades 104 are fixed on the fairing 103. A fixing bolt 105 is provided between the blades 104 and the fairing 103. The multi-dimensional sensing module 2 includes an acoustic fingerprint sensor 201, an ultrasonic bolt axial force detection sensor 202, and a temperature and vibration integrated sensor 203. The acoustic fingerprint sensor 201 is installed inside the blades 104 and on the tower base 101, the ultrasonic bolt axial force detection sensor 202 is installed on the fixing bolt 105, and the temperature and vibration integrated sensor 203 is installed inside the blades 104.

[0043] The acoustic signature sensor 201 includes two types. The tower base 101 uses a free-field acoustic signature sensor 201: a 1 / 2-inch pre-polarized condenser microphone with a frequency response range of 10Hz-40kHz and a sensitivity of -40dB±2dB, installed in the open area at the front of the tower base 101. The blade 104 uses a pressure-field acoustic signature sensor 201: a 1 / 4-inch pre-polarized condenser microphone with a frequency response range of 20Hz-20kHz and a sensitivity of -30dB±2dB, embedded in the sealed cavity of the blade 104. 1-3 temperature and vibration integrated sensors 203 are deployed on each blade 104. These sensors use domestic MEMS chips, have a frequency response range of 0Hz-10kHz, a temperature range of -40℃ to 125℃, and an accuracy of ±0.5℃. One sensor is deployed at the root, tip, and middle main beam of each blade 104. Ultrasonic bolt axial force monitoring sensors are deployed on the surface of each fixing bolt 105 of each fan blade 104. These sensors are based on piezoelectric ultrasonic technology, with a center frequency of 2MHz±0.2MHz and a bandwidth of ≥1MHz.

[0044] The edge computing module is deployed in an edge computing box at the bottom of the wind turbine tower. It includes a data preprocessing unit and a lightweight AI inference unit. The data preprocessing unit is used to perform adaptive spectral subtraction denoising and angular domain resampling on the acoustic signal; to perform baseline calibration on the temperature and vibration signal; and to perform temperature compensation on the bolt axial force signal. The lightweight AI inference unit deploys a lightweight model to identify in real time the sudden drop in axial force of the fixing bolt 105 and the abnormal acceleration peak of the blade 104 exceeding the threshold and triggering alarms.

[0045] The cloud-based AI module is deployed in the wind farm monitoring center and includes a multi-source data fusion unit, a deep diagnostic model unit, and an operation and maintenance decision unit. The multi-source data fusion unit is used to integrate acoustic signature, temperature and vibration, and bolt axial force data. The deep diagnostic model unit uses a transfer learning algorithm, combined with a convolutional neural network to extract acoustic signature features and a random forest to assess the severity of defects.

[0046] Furthermore, it also includes an air blowing dust removal mechanism 3, which includes a sealing cover 301, a winding drum 305, a drive motor 307, a high-pressure air pump 317, and an air blowing airbag 316. The sealing cover 301 is fixedly installed inside the tower base 101. The acoustic sensor 201 is installed at the front end of the sealing cover 301. The winding drum 305 is installed inside the sealing cover 301 and rotatably connected to it. The drive motor 307 is installed inside the sealing cover 301 to drive the winding drum 305. The air blowing airbag 316 is installed on the winding drum 305. The high-pressure air pump 317 is installed inside the sealing cover 301 to supply air to the air blowing airbag 316.

[0047] The airbag 316 is rolled up and unrolled by the drive motor 307. When it is necessary to clean the voiceprint sensor 201, the airbag 316 is unfolded, and then air is supplied by the high-pressure air pump 317. The airbag 316 blows the speaker unit 2012 inside the voiceprint sensor 201, blowing out the sand and dust inside the protective cover 2011, so as to avoid the sand and dust affecting the sensitivity of the voiceprint sensor 201 and thus affecting the detection accuracy.

[0048] Furthermore, the front end face of the sealing cover 301 is provided with an equipment compartment 302, and the winding drum 305, the drive motor 307 and the high-pressure air pump 317 are all installed in the equipment compartment 302. The side end face of the equipment compartment 302 is provided with a guide groove 308, and the top of the equipment compartment 302 is provided with a sealing cover 303 that slides with the guide groove 308.

[0049] The device is installed in the equipment compartment 302. A sliding sealing cover 303 structure is provided on the top of the equipment compartment 302. When the acoustic sensor 201 needs to be cleaned, the sealing cover 303 is opened to carry out the cleaning operation. After the cleaning operation is completed, the sealing cover 303 is closed, thereby preventing dust or water from entering the equipment compartment 302 and affecting the service life of the device.

[0050] Furthermore, the drive motor 307 is fixedly installed inside the equipment compartment 302, a rack 309 is fixedly installed on the lower end face of the sealing cover 303, and a gear 310 is fixedly connected on the output shaft of the drive motor 307.

[0051] The drive motor 307 drives the gear 310 to rotate, thereby the gear 310 and the rack 309 work together to drive the opening and closing of the sealing cover 303, so that the opening and closing of the sealing cover 303 moves in coordination with the rolling and unfolding of the airbag 316.

[0052] Furthermore, a bearing seat 306 is provided inside the equipment compartment 302. The bearing seat 306 is provided with a first shaft hole 3161 that is rotatably connected to the take-up drum 305. A first synchronous pulley 311 is provided on the output shaft of the drive motor 307. A second synchronous pulley 312 is provided on the side end face of the take-up drum 305. A synchronous belt 313 for transmission is provided on the first synchronous pulley 311 and the second synchronous pulley 312.

[0053] The drive motor 307 drives the first synchronous pulley 311 to rotate, which in turn drives the second synchronous pulley 312 and the synchronous belt 313 to rotate the take-up drum 305, so that the drive motor 307 and the synchronous belt 313 drive the take-up drum 305 and the sealing cover 303 to move.

[0054] Furthermore, the high-pressure air pump 317 is fixedly installed inside the equipment compartment 302. The air outlet end of the high-pressure air pump 317 is provided with an air supply pipe 315. The end of the air supply pipe 315 is provided with an air supply cover 314 that cooperates with the take-up drum 305. The take-up drum 305 is provided with a first air hole 3151 that communicates with the air supply cover 314. The top of the take-up drum 305 is provided with a second air hole 3152 that communicates with the air bladder 316. The side end face of the air supply cover 314 is provided with a second shaft hole 3141 that cooperates with the take-up drum 305.

[0055] The high-pressure air pump 317 inflates the inside of the take-up drum 305 through the air supply pipe 315. The airflow flows through the take-up drum 305 to the air bladder 316. The air bladder (316) inflates under the action of air pressure until it is aligned with the acoustic sensor 201. Then it blows the acoustic sensor 201. The rotatably connected take-up drum 305 and air supply cover 314 cooperate to ensure that the take-up drum 305 can stably supply air during rotation.

[0056] Furthermore, the airbag 316 has a circular structure, and the bottom of the airbag 316 is provided with an inflation hole 3163 that communicates with the winding drum 305. The voiceprint sensor 201 includes a protective cover 2011 and a speaker unit 2012. The protective cover (2011) is provided with a through hole 2013, and the airbag 316 is uniformly provided with air heads 3161 corresponding to the positions of the through holes 2013.

[0057] The air inlet 3161 on the airbag 316 is aligned with the through hole 2013 on the protective cover 2011, so that the airflow is blown directly into the protective cover 2011 to clean the sand and dust inside the protective cover 2011.

[0058] Furthermore, a rain shield 304 is provided on the front end face of the sealing cover 301, a magnet 3041 is provided on the lower end face of the rain shield 304, and an iron piece 3162 corresponding to the position of the magnet 3041 is provided on the top of the airbag 316.

[0059] The top of the acoustic sensor 201 is covered by the rain shield 304 to prevent rainwater from entering the acoustic sensor 201 and affecting its service life. An iron plate 3162 is set inside the air bladder 316 so that after the air bladder 316 is opened, the top can be attracted to the rain shield 304 by the iron plate 3162, preventing the air bladder 316 from shaking, so that the air head 3161 can be accurately aligned with the through hole 2013.

[0060] Furthermore, there are three integrated temperature and vibration sensors 203 inside the blade 104, with the three sets of integrated temperature and vibration sensors 203 located at the top, middle and bottom of the blade 104, respectively.

[0061] Three sets of sensors are deployed at the root, tip, and middle main beam of each blade 104, one of each, to synchronously collect vibration and temperature signals of the blade 104 and monitor anomalies such as icing and hot spots.

[0062] When dust removal is required for the acoustic sensor 201, the drive motor 307 starts, and the gear 310 on its output shaft rotates. Through its engagement with the rack 309 on the lower end face of the sealing cover 303, the sealing cover 303 slides open along the guide groove 308 on the side end face of the equipment compartment 302. At the same time, the first synchronous pulley 311 on the output shaft of the drive motor 307 drives the second synchronous pulley 312 on the take-up drum 305 to rotate via the synchronous belt 313. This causes the take-up drum 305 to rotate and expand the air bladder 316. The expanded air bladder 316 inflates under the action of air supplied by the high-pressure air pump 317 through the air supply pipe 315, the air supply cover 314, and the first air hole 3151 and the second air hole 3152 on the take-up drum 305. The top iron plate 3162 is attracted to the magnet 3041 on the lower end of the rain shield 304 at the front end of the sealing cover 301 to maintain stability. At this time, the air blowing head 3161 on the air blowing bag 316 corresponds one-to-one with the through hole 2013 on the protective cover 2011 of the acoustic sensor 201. The high-pressure airflow is blown into the protective cover 2011 through the air blowing head 3161 to blow out the sand and dust inside. After the dust removal is completed, the high-pressure air pump 317 stops supplying air, the air blowing bag 316 contracts, the drive motor 307 reverses, and drives the winding drum 305 to rotate in the opposite direction through the synchronous belt 313 to wind up the air blowing bag 316. At the same time, the gear 310 and the rack 309 cooperate to drive the sealing cover 303 to slide and close along the guide groove 308, sealing the equipment compartment 302.

[0063] Example 2

[0064] Please see Figures 1-15 A method for detecting defects in wind turbine blades using multi-dimensional sensing and AI, comprising the following steps:

[0065] Step 1: Multi-dimensional data acquisition and deployment of multiple types of sensors to construct a full-link sensing network for blade 104: Acoustic sensor 201 uses a 1 / 2-inch pre-polarized condenser microphone with a frequency response of 10Hz-40kHz and a sensitivity of -40dB±2dB. One microphone is installed at the bottom of the tower to capture wind noise events of blade 104, such as crack friction noise and bolt breakage noise. Three 1 / 4-inch pressure field acoustic sensors 201 are embedded in the internal web / beam cap cavity of blade 104, with a frequency response of 20Hz-20kHz and a sensitivity of -30dB±2dB, to monitor internal structural damage and pressure fluctuations. An ultrasonic bolt axial force detection sensor is also included. Device 202 is based on piezoelectric ultrasonic technology, with a center frequency of 2MHz±0.2MHz and a bandwidth of ≥1MHz. One device is deployed on the surface of each wind turbine blade root bolt. The preload is calculated by measuring the ultrasonic wave propagation time difference and combining it with a temperature compensation algorithm. Temperature and vibration integrated sensor 203 uses domestic MEMS chips. One sensor is deployed at the root, tip, and middle main beam of each blade. It synchronously collects vibration acceleration and temperature. All sensors are time-stamp aligned through a synchronous clock and edge computing nodes. The data is transmitted in real time to the edge computing box via gigabit Ethernet, forming a multi-dimensional raw data stream containing acoustic signature, temperature and vibration, and bolt axial force.

[0066] Step 2: Real-time preprocessing of the edge layer. The original blade 104 operating sound signal picked up by the microphone contains wind noise.

[0067] Various occasional interfering noises such as yaw, pitch, drain hole, bird calls, and sheep bleating can be effectively separated by background noise removal, thus effectively isolating the target sweeping sound events generated by the operation of blade 104. Based on the short-time logarithmic energy variation characteristics of the blade 104 rotational sound signal, the instantaneous angular velocity of the impeller is calculated. Using this, the non-stationary time-domain sweeping sound signal is resampled at the same angle, and a stationary angular-domain sweeping sound signal is fitted to eliminate the influence of rotational speed fluctuations on subsequent extraction of damage abnormal sound acoustic features. Since surface damage alters the spectral energy distribution of the blade 104 sweeping sound signal and its characteristics are perceptible and distinguishable by the human ear, the frequency decomposition characteristics of the cochlea are simulated. Based on the frequency band energy distribution and concentration of irregular damage abnormal sounds... Based on the degree and proportion of the frequency band, the number, weight and amplitude of the Mel filter are adaptively adjusted. A frequency band adaptive sensing spectrogram enhancement representation method for wind-swept sound signals is proposed to avoid the loss or masking of key frequency band representation information caused by auditory masking characteristics. By analyzing the mapping relationship between temporal voiceprint feature points and damaged abnormal sounds in the enhanced spectrogram, as well as the gradient change law of the neighborhood cepstral energy value of damaged feature pixels, a variable-scale differential strategy is designed to adaptively adjust the differential direction and interval between pixels, stabilize the neighborhood gradient change trend of voiceprint feature points, and obtain dynamic voiceprint feature map. Together with the static feature map, a dynamic and static dual-stream voiceprint feature map is established to characterize the abnormal sounds of damage on irregular surfaces.

[0068] In its free state, a bolt has no preload. However, in its tightened state, due to the preload, the bolt deforms, resulting in a deformation of ΔL. The measuring system calculates the preload F based on the mathematical relationship between ΔL and the preload F: F = (E * S * ΔL) / L, where F is the bolt's preload; E is the elastic modulus of the bolt material; S is the bolt's cross-sectional area; ΔL is the bolt's deformation; and L is the clamping length of the bolt assembly. Using the formula F = (E * S * ΔL) / L, the measuring system calculates the current preload F of the bolt based on ΔL. The preload measurement system transmits and receives ultrasonic pulse electrical signals, measures and calculates the time difference between the transmitted and echo electrical signals. When the bolt is in a free state, the time difference between the transmitted and received electrical signals is T0. When the bolt is in a tightened state, the time difference between the transmitted and received electrical signals is T1. Based on the relationship between the time difference between the transmission and reception of electrical signals and the deformation of the bolt, the deformation of the bolt is obtained as ΔL = 1 / 2(T1-T0)·v, where v is the propagation speed of the mechanical longitudinal wave in the bolt. Finally, the measurement system can obtain the preload of the bolt in the current state based on ΔL and the formula F = (E*S*ΔL) / L.

[0069] Step 3: Preliminary edge layer diagnosis. Deploy a lightweight AI model (based on an improved MobileNet-V3, model size <50MB) to achieve real-time analysis. Input preprocessed acoustic signature (Mel spectrum), temperature and vibration (time domain + frequency domain features), and bolt axial force (time series) data. Extract local features (such as abrupt changes in high-frequency energy of the acoustic signature and abnormal temperature and vibration kurtosis) through convolutional layers. Output the discrimination results of sudden drop in bolt axial force, abnormal vibration of blade 104, and temperature exceeding the threshold through fully connected layers. If an alarm is triggered, the edge node pushes an audible and visual alarm to the operation and maintenance platform through the 4G / 5G network and caches the original data to the local SSD. Non-abnormal data is stored in the edge storage array after hash verification for subsequent model training data supplementation.

[0070] Step 4: Cloud-based deep diagnostics. Integrating multi-source preprocessed data to generate health feature vectors, a transfer learning model (source domain: 5 years of historical fault data; target domain: current wind field data) is used to align the distribution differences between the source and target domains via Grassmann manifolds, addressing the problem of scarce cross-wind field samples. A CNN sub-network extracts deep acoustic features, capturing the Mel frequency shift of 0.5mm microcracks. A random forest sub-network fuses temperature and vibration data with bolt data to assess defect severity. Input features include historical crack propagation rate of 0.1mm / month, current temperature and humidity, and bolt axial force loss rate. Finally, an LSTM model predicts the remaining service life (RUL), with inputs including defect size, environmental parameters (temperature, humidity, salt spray concentration), and historical propagation rate. The output RUL error is ≤15%.

[0071] Step 5: Operation and maintenance decision-making and closed loop. Based on the diagnostic results, a three-level early warning is generated: Level 1: Immediate shutdown, triggered by bolt axial force loss > 15% or crack length > 3mm, pushing to the SCADA system to force load reduction and lock the wind turbine; Level 2: Maintenance within 48 hours, generating a work order containing defect location, type, and RUL prediction value, the operation and maintenance platform optimizes the inspection path through the Dijkstra algorithm, increasing the inspection frequency of high-risk blades 104 by 50%; Level 3: Planned maintenance, included in the long-term monitoring database, recording defect development data for model iteration.

[0072] The above are merely preferred embodiments of the present invention and are not intended to limit the present invention. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A wind turbine blade defect detection method combining multidimensional sensing and AI, comprising a wind turbine generator body (1), characterized in that: It also includes a multi-dimensional sensing module (2), an edge computing module, and a cloud AI module; The wind turbine main body (1) includes a tower base (101), a generator head (102), a fairing (103), and blades (104). The generator head (102) is installed on the top of the tower base (101), the fairing (103) is installed at the front end of the generator head (102), and the blades (104) are fixed on the fairing (103). A fixing bolt (105) is provided between the blades (104) and the fairing (103). The multi-dimensional sensing module (2) includes an acoustic fingerprint sensor (201), an ultrasonic bolt axial force detection sensor (202), and a temperature and vibration integrated sensor (203). The acoustic fingerprint sensor (201) is installed inside the blades (104) and on the tower base (101). The ultrasonic bolt axial force detection sensor (202) is installed on the fixing bolts (105), and the temperature and vibration integrated sensor (203) is installed inside the blades (104).

2. The wind turbine blade defect detection method combining multi-dimensional sensing and AI as described in claim 1, characterized in that: It also includes an air blowing dust removal mechanism (3), which includes a sealing cover (301), a winding drum (305), a drive motor (307), a high-pressure air pump (317), and an air blowing airbag (316). The sealing cover (301) is fixedly installed inside the tower base (101). The acoustic sensor (201) is installed at the front end of the sealing cover (301). The winding drum (305) is installed inside the sealing cover (301) and rotatably connected to it. The drive motor (307) is installed inside the sealing cover (301) to drive the winding drum (305). The air blowing airbag (316) is installed on the winding drum (305). The high-pressure air pump (317) is installed inside the sealing cover (301) to supply air to the air blowing airbag (316).

3. The wind turbine blade defect detection method combining multi-dimensional sensing and AI as described in claim 2, characterized in that: The front end face of the sealing cover (301) is provided with an equipment compartment (302). The winding drum (305), drive motor (307) and high-pressure air pump (317) are all installed in the equipment compartment (302). The side end face of the equipment compartment (302) is provided with a guide groove (308). The top of the equipment compartment (302) is provided with a sealing cover (303) that slides with the guide groove (308).

4. The wind turbine blade defect detection method combining multi-dimensional sensing and AI as described in claim 3, characterized in that: The drive motor (307) is fixedly installed inside the equipment compartment (302), and a rack (309) is fixedly installed on the lower end face of the sealing cover (303). A gear (310) is fixedly connected to the output shaft of the drive motor (307).

5. The wind turbine blade defect detection method combining multi-dimensional sensing and AI as described in claim 4, characterized in that: The equipment compartment (302) is provided with a bearing seat (306), and the bearing seat (306) is provided with a first shaft hole (3161) that is rotatably connected to the winding drum (305). The output shaft of the drive motor (307) is provided with a first synchronous pulley (311), and the side end face of the winding drum (305) is provided with a second synchronous pulley (312). The first synchronous pulley (311) and the second synchronous pulley (312) are provided with a synchronous belt (313) for transmission.

6. The wind turbine blade defect detection method combining multi-dimensional sensing and AI as described in claim 5, characterized in that: The high-pressure air pump (317) is fixedly installed in the equipment compartment (302). The air outlet of the high-pressure air pump (317) is provided with an air supply pipe (315). The end of the air supply pipe (315) is provided with an air supply cover (314) that cooperates with the winding drum (305). The winding drum (305) is provided with a first air hole (3151) that communicates with the air supply cover (314). The top of the winding drum (305) is provided with a second air hole (3152) that communicates with the air bladder (316). The side end face of the air supply cover (314) is provided with a second shaft hole (3141) that cooperates with the winding drum (305).

7. The wind turbine blade defect detection method combining multi-dimensional sensing and AI as described in claim 2, characterized in that: The airbag (316) has a circular structure. The bottom of the airbag (316) is provided with an inflation hole (3163) that communicates with the winding drum (305). The voiceprint sensor (201) includes a protective cover (2011) and a speaker unit (2012). The protective cover (2011) is provided with a through hole (2013). The airbag (316) is uniformly provided with air blowing heads (3161) corresponding to the positions of the through holes (2013).

8. The wind turbine blade defect detection method combining multi-dimensional sensing and AI as described in claim 2, characterized in that: The front end face of the sealing cover (301) is provided with a rain shield (304), the lower end face of the rain shield (304) is provided with a magnet (3041), and the top of the airbag (316) is provided with an iron piece (3162) corresponding to the position of the magnet (3041).

9. The wind turbine blade defect detection method combining multi-dimensional sensing and AI as described in claim 2, characterized in that: The blade (104) contains three integrated temperature and vibration sensors (203), with the three sets of integrated temperature and vibration sensors (203) located at the top, middle and bottom of the blade (104), respectively.

10. A method for detecting defects in wind turbine blades (104) combining multidimensional sensing and AI, applicable to the wind turbine blade defect detection method combining multidimensional sensing and AI as described in any one of claims 1-9, characterized in that, Includes the following steps: Step 1: Multidimensional data acquisition, deploying acoustic fingerprint sensor (201), ultrasonic bolt axial force detection sensor (202) and temperature and vibration integrated sensor (203) to acquire blade (104) data in real time; Step 2: Real-time preprocessing of the edge layer, adaptive spectral subtraction denoising of the acoustic signal, baseline calibration and extraction of time-domain features of the temperature and vibration signal, and temperature compensation correction of the bolt axial force signal. Step 3: Preliminary diagnosis at the edge layer. The preprocessed data is analyzed using a lightweight AI model to identify sudden drops in bolt axial force and abnormal vibration of blades (104) and trigger alarms. Non-abnormal data is stored locally. Step 4: Cloud-based deep diagnostics, integrating multi-source preprocessed data to generate health feature vectors, using a transfer learning model to identify defect types and locations, and combining a random forest model to predict remaining service life; Step 5: Operation and maintenance decision-making and closed loop. Generate early warning information and push it to the operation and maintenance platform. Optimize the inspection plan and record defect development data for model iteration.

Images